But a small collective running that box, especially spanning timezones, could potentially be a viable alternative or will be soon -- with obv privacy gains too
Conversely, it's useful to get an immediate answer sometimes
6 months ago, "what temp is pork safe at?" was a few clicks, long SEO optimised blog post answers and usually all in F not C ... despite Google knowing location ... I used it as an example at the time of 'how hard can this be?'
First sentance of Google AI response right now: "Pork is safe to eat when cooked to an internal temperature of 145°F (63°C)"
Dear lord please don’t use an AI overview answer for food safety.
If you made a bet with your friend and are using the AI overview to settle it, fine. But please please click on an actual result from a trusted source if you’re deciding what temperature to cook meat to
The problem is that SEO has made it hard to find trustworthy sites in the first place. The places I trust the most now for getting random information is Reddit and Wikipedia, which is absolutely ridiculous as they are terrible options.
But SEO slop machines have made it so hard to find the good websites without putting in more legwork than makes sense a lot of the time. Funnily enough, this makes AI look like a good option to cut through all the noise despite its hallucinations. That's obviously not acceptable when it comes to food safety concerns though.
If I do that search on Google right now, the top result is the National Pork Board (pork.org): ad-free, pop-up free, waffle-free and with the correct answer in large font at the top of the page. It's in F, but I always stick " C" at the end of temperature queries. In this case that makes the top result foodsafety.gov which is equally if not more authoritative, also ad-, waffle-, and popup- free and with with the answer immediately visible.
Meanwhile the AI overview routinely gives me completely wrong information. There's zero chance I'm going to trust it when a wrong answer can mean I give my family food poisoning.
I agree that there is a gigaton of crap out there, but the quality information sources are still there too. Google's job is to list those at the top and it actually has done so this time, although I'll acknowledge it doesn't always and I've taken to using Kagi in preference for this reason. A crappy AI preview that can't be relied on for anything isn't an acceptable substitute.
Kagi search gives the pork board first as well. But note that site fails mtkd's requirements giving temperature in degrees Fahrenheit and not Celsius. The second hit does give a correct temperature but has a cookie banner (which at least can be rejected with one click)
The optional Kagi assistant quotes the pork board, usda which also is only in Fahrenheit, and third a blog on site for a thermometer that quoted the UK Food Standard Authority and gives its temperature
However there is a problem the UK FSA does not agree with USDA on the temperature it puts it higher at 70 degrees C rather than 63
So if you get the USDA figure you are taking a risk.
The Kagi Assistant gives both temperatures but it is not clear which one is correct although both figures are correctly linked to the actual sites.
I don't really see the problem with F and C. As I mentioned, I always stick " C" on the end of temperature queries. It's 2 characters and the results always have the centigrade temps, on both Kagi and Google.
Of course. What else would I think they were complaining about? I also live in a country that uses C. That's why I always stick " C" on the end of temperature queries.
It would be nice if they automatically prioritised those results, but that's a search engine improvement and nobody's working on those any more [1]. A half-arsed AI summary that can't be trusted to get the actual temperature right certainly doesn't solve it.
[1] Except Kagi, and even they're distracted by the AI squirrel.
As I said, it routinely gives incorrect data so it can't be relied on for something that matters, like a safe cooking temperature.
Note that we're talking about the Google AI Summary here, not AI in general. Whatever magical capabilities you think your favoured model has or will soon have, the Google AI Summary is currently utter garbage and routinely spouts nonsense. (Don't try and persuade me otherwise. I have to use Google at work so I see its lies first hand every day.)
It's a bit like the Easter Islanders cutting down all of their trees for wood. Where does Google management think they'll get search results if they kill the entire internet? Has anyone at Google thought that far ahead?
The internet they dream of is like a large mall. It consists of service providers selling you something, and Google directing you to them in exchange for some of the profit. The role of users in this model is that of a Piñata that everyone hits on to drop some money.
DDG is just serving you remixed bing and yandex results. There’s basically no alternative to GBY that do their own crawling and maintain their own index.
I've been using noai.duckduckgo.com for a few weeks now and it's pretty reliable. Still yandex etc. but at least no longer AI overview. (Yes, I know about settings, but they get deleted every restart).
Well, they decided which sites were OK, and then people SEO'd a bunch of crap into Google's idea of a good website.
I'm no fan of Google but it's not so simple to figure out what's relevant, good content on the scale of the internet, while confronted by an army of adversarial actors who can make money by working out what you value in a site.
AI is being influenced by all that noise. It isn’t necessarily going to an authoritative source, it’s looking at Reddit and some SEO slop and using that to come up with the answer.
We need AI that’s trained exclusively on verified data and not random websites and internet comments.
I asked Gemini about some Ikea furniture dimensions and it gave seemingly correct answers, until it suddenly didn't make sense.
Turns out all the information it gave me came from old Reddit posts and lots of it was factually wrong. Gemini however still linked some official Ikea pages as the "sources".
It'll straight up lie to you and then hide where it actually got it's info from. Usually Reddit.
Creating better datasets would also help to improve the performance of the models, I would assume. Unfortunately, the costs to produce high-quality datasets of a sufficient size seem prohibitive today.
I'm hopeful this will be possible in the future though, maybe using a mix of 1) using existing LLMs to help humans filter the existing internet-scale datasets, and/or 2) finding some new breakthroughs to make model training more data efficient.
Please find a trusted source of information for food safety information.
It's genuinely harder than it's ever been to find good information on the internet, but when you're dealing with food safety information, it's really worth taking the extra minute to find a definitive source.
The site you linked to is published by the American government who are actively anti-science, removing access to research and cutting funding for ideological reasons, denying widely accepted climate and medical facts etc. I wouldn't trust that source at all.
I mean, I'm absolutely not going to take the current FDA's guidance on drinking raw milk. You need to evaluate what you're willing to trust for yourself. I printed out the pre-January 2025 FDA vaccine schedule, so I could compare it against any updated recommendations.
All that said, the cooking temperatures aren't something that they've changed, and I would still rely on the US government published number over a AI generated value
Mmm, I see this cutting both ways -- generally, I'd agree; safety critical things should not be left to an AI. However, cooking temperatures are information that has a factual ground truth (or at least one that has been decided on), has VERY broad distribution on the internet, and generally is a single, short "kernel" of information that has become subject to slop-ifying and "here's an article when you're looking for about 30 characters of information or less" that is prolific on the web.
So, I'd agree -- safety info from an LLM is bad. But generally, the /flavor/ (heh) of information that such data comprises is REALLY good to get from LLMs (as opposed to nuanced opinions or subjective feedback).
I don’t know. I searched for how many chapters a popular manga has on Google and it gave me the wrong answer (by an order of magnitude). I only found out later and it did really piss me off because I made a trek to buy something that never existed. I should’ve known better.
I don’t think this is substantively different from cooking temperature, so I’m not trusting that either.
Eh I think it is. Arcane things -- sure, that might be a bit of a stretch. My general rule of thumb is that if I would expect ~10% of people to know the information factually, I can likely trust what an LLM tells me.
Idk. Maybe that's true today (though even today I'm not sure) but how long before AI becomes better than just finding random text on a website?
After all, AI can theoretically ask follow-up questions that are relevant, can explain subtleties peculiar to a specific situation or request, can rephrase things in ways that are clearer for the end user.
Btw, "What temperature should a food be cooked to" is a classic example of something where lots of people and lots of sources repeat incorrect information, which is often ignored by people who actually cook. Famously, the temp that is often "recommended" is only the temp at which bacteria/whatever is killed instantly - but is often too hot to make the food taste good. What is normally recommended is to cook to a lower temperature but keep the food at that temperature for a bit longer, which has the same effect safety-wise but is much better.
> Btw, "What temperature should a food be cooked to" is a classic example of something where lots of people and lots of sources repeat incorrect information, which is often ignored by people who actually cook. Famously, the temp that is often "recommended" is only the temp at which bacteria/whatever is killed instantly
I love this reply because you support your own point by repeating information that is technically incorrect.
To qualify myself, I have a background in food service. I've taken my "Food Safe" course in Ontario which is not legally mandated to work in food service, but offered by our government-run health units and many restaurants require a certificate to be employed in any food handling capacity (wait staff or food prep).
There is no such thing as "killed instantly." The temperature recommendations here in Canada, for example, typically require that the food be held at that temperature for a minimum of 15 seconds.
There is some truth in what you say. Using temperature to neutralize biological contaminants is a function of time and you can certainly accomplish the same result by holding food at lower temperature for a longer period of time. Whether this makes the food "taste better" or not depends on the food and what you're doing.
Sous Vide cooking is the most widely understood method of preparation where we hold foods at temperatures that are FAR lower than what is typically recommended, but held for much longer. I have cooked our family Thanksgiving Turkey breast at 60C sous vide, and while I personally like it... others don't like the texture. So your mileage may vary.
My point is that you're making a bunch of claims that have grains of truth to them, but aren't strictly true. I think your comment is an application of the dunning kruger effect. You know a little bit and because of that you think you know way more than you actually do. And I had to comment because it is beautifully ironic. Almost as if that paragraph in your comment is, itself, AI slop lol
> And I had to comment because it is beautifully ironic. Almost as if that paragraph in your comment is, itself, AI slop lol
Glad to be of service :) I think that's the first time a comment of mine has been accused of being AI slop. Sorry to say, every word - correct or incorrect - is mine.
> To qualify myself, I have a background in food service.
I'm just a person who watches cooking YouTube a bit, so right off the bat - I'll defer to your expertise on this.
I'm not sure we really disagree much though. My rough memory is that the guidelines specify temperature at which things are killed, I don't know if "instantly" or "15 seconds" really makes a difference in practice.
> Sous Vide cooking is the most widely understood method of preparation where we hold foods at temperatures that are FAR lower than what is typically recommended, but held for much longer.
Sous Vide is where I was first exposed to this concept, but I was more referring to things like chicken breasts, etc, which often aren't great at the minimal internal temperature, but I've seen YouTube "chefs" recommend cooking them to a slightly lower temperature, banking on the idea that they will effectively be at a slightly lower temperature, but long enough to still effectively kill bacteria. I've even seen criticism of the FDA charts for exactly this reason.
But to clarify, this is far outside any expertise I actually have.
> I think your comment is an application of the dunning kruger effect. You know a little bit and because of that you think you know way more than you actually do.
> , but I was more referring to things like chicken breasts, etc, which often aren't great at the minimal internal temperature, but I've seen YouTube "chefs" recommend cooking them to a slightly lower temperature, banking on the idea that they will effectively be at a slightly lower temperature, but long enough to still effectively kill bacteria.
Something that is very common is removing food from the heat source before the internal temperature hits your target because heat transfer itself takes time and so the food will continue to cook inside for a short period of time after being removed. This is because the outside, which you are blasting with heat that is far higher than your internal target, will partly transfer to the inside (the rest will dissipate into the air). So if you remove it from the heat exactly when you hit the internal temp, you can exceed the target temperature and your food will be "over cooked."
The problem with a tv chef recommending using a traditional cooking method, such as baking or frying, to TARGET a lower temperature, is that is very hard with those mediums to control for time. What you are doing with those mediums is you are blasting the outside of your food with a temperature that is far hotter than your internal target.
And so say, for example, you have your oven set to 180C and you are cooking chicken and your internal temperature target is, let's say 4 degrees cooler than the official recommendation. So the official recommendation is 74C held for a minimum of 15 seconds (that's Canada) and you are targeting 70C. With traditional cooking methods, you are playing a guessing game where you're blasting the outside of the food with very hot temperatures in order to bring the inside of the food up to some target.
I don't know off hand how much longer you would have to hold at 70C to get the same effect as 15 seconds at 74C ... but while you're waiting, your food is likely to exceed 74C anyway because of the high temperatures you're working with.
So that's why I talked about sous vide... becuase it's kind of the only way you can control for those variables. No oven can hold steady at temps as low as 70C (even at higher temps they fluctuate quite a bit. Anywhere from 5C - 20C depending on the oven).
And yeah - we definitely agree on most things. The minimum recommended temperatures are "play it safe" rather than "make food delicious." I do recognize that that was ultimately your point :)
It wasn't really my point to pick on you or argue with you, but to show that certain things you said are "partly true", which is a common complaint of AI (that and hallucinations). When we're dealing with things like food safety and the general public, it is usually better to offer general advise that is play it safe.
And with certain foods this matters more than others. Chickens get infected with salmonella while they are alive, for example, and so the bacteria can live throughout the meat. Whereas if you're cooking beef, you really only need to worry about surface contamination and so you can sear a steak for a few seconds and have it still be very "blue" in the middle and you're good.
When I searched for the safe temperature for pork (in German), I found this as the first link (Kagi search engine)
> Ideally, pork should taste pink, with a core temperature between 58 and 59 degrees Celsius. You can determine the exact temperature using a meat thermometer.
Is that not a health concern?
Not anymore, as nutrition expert Dagmar von Cramm confirms:
> “Trichinae inspection in Germany is so strict — even for wild boars — that there is no longer any danger.”
I was just thinking that EU sources might be a good place to look for this sort of thing, given that we never really know what basic public health facts will be deemed political in the US on any given day. But, this reveals a bit of a problem—of course, you guys have food safety standards, so advice they is safe over there might not be applicable in the US.
Doesn't even have to be "better", just "different". The classic one is whether you should refrigerate eggs, which has diametrically opposite answers.
But anything that actually matters could be politicized at any time. I remember the John Gummer Burger Incident: http://news.bbc.co.uk/1/hi/uk/369625.stm , in the controversy over whether prion diseases in beef (BSE) were a problem.
Funny story, I used that to know the cooked temperature of burgers, it said medium-rare was 130. I proceeded to eating it and all, but then like half way through, I noticed the middle of this burger is really red looking, doesn't seem normal, and suddenly I remembered, wait, ground beef is always supposed to be 160, 130 medium-rare is for steak.
I then chatted that back to it, and it was like, oh ya, I made a mistake, you're right, sorry.
Anyways, luckily I did not get sick.
Moral of the story, don't get mentally lazy and use AI to save you the brain it takes for simple answers.
Do you actually put a thermometer in your burgers/steaks/meat when you’re cooking? That seems really weird.
Why are people downvoting this? I’ve literally never seen anyone use a thermometer to cook a burger or steak or pork chop. A whole roasted turkey, sure.
You're getting lots of thermometer answers, so I'm going to give the opposite: I'm also on team "looks good to me" + "cooking time on packet" + "just cut it and look"
Many people wing dishes that they've prepared 100s of times. Others rarely make the same recipe twice. Neither are correct or incorrect, but the latter is very much going to measure everything they're doing carefully (or fail often).
For whole meats, it's usually safe to be rare and you can tell that by feel, though a thermometer is still useful if you aren't a skilled cook or you are cooking to a doneness you aren't familiar with.
I think your reference pool is just small. I absolutely use it for meat and especially for ground meat, which has a much higher chance of contamination.
Perhaps this varies by region? I don't know anyone that buys burgers in a packet. They buy ground beef and either make patties or balls (for smash burgers).
I don't do much of the shopping, but we get costco frozen burger patties for most of our home burgers. I don't think it costs more than the same weight of 'whole' ground beef, and it's convenient.
Those are thin enough I wouldn't think to stick a thermometer in them... it would be too hard to get it in the center and not out the other side, and it's pretty easy to get a sense of doneness from the outside (or cut into one and see). Steaks, depending on who's eating and doneness preferences, thermometer is nice. Roasts, almost certainly.
Why would you purchase meat that you suspect is diseased? Even if you cook it well-done, all the (now dead) bacteria and their byproducts are still inside. I don't understand why people do this to themselves? If I have any suspicion about some meat, I'll throw it away. I'm not going to cook it.
Most of the time if there isn't a straightforward primary source in the top results, Google's AI overview won't get it right either.
Given the enormous scale and latency constraints they're dealing with, they're not using SOTA models, and they're probably not feeding the model 5000 words worth of context from every result on the page.
Not only that, it includes a link to the USDA reference so you can verify it yourself. I have switched back to google because of how useful I find the RAG overviews.
The link is the only useful part, since you can’t trust the summary.
Maybe they could just show the links that match your query and skip the overview. Sounds like a billion-dollar startup idea, wonder why nobody’s done it.
It’s a pretty good billion dollar idea, I think you’ll do well. In fact I bet you’ll make money hand over fist, for years. You could hire all the best engineers and crush the competition. At that point you control the algorithm that everyone bases their websites on, so if you were to accidentally deploy a series of changes that incentivized low quality contentless websites… it wouldn’t matter at all; not your problem. Now that the quality of results is poor, but people still need their queries answered, why don’t you provide them the answers yourself? You could keep all the precious ad revenue that you previously lost when people clicked on those pesky search results.
As of a couple weeks ago it had a variety of unsafe food recommendations regarding sous vide, e.g. suggesting 129F for 4+ hours for venison backstrap. That works great some of the time but has a very real risk of bacterial infiltration (133F being similar in texture and much safer, or 2hr being a safer cook time if you want to stick to 129F).
The problem is that 129F is a borderline safe temperature at any length of time. 3-4hrs is closer to what you want to kill _all_ the salmonella and e coli, but 2hrs kills enough that it's very likely to be safe, and 129F isn't a sufficient temperature for any length of time to kill clostridium perfringens and other similar bacteria, resulting in spoilage occasionally with a greatly increased likelihood the longer you cook it.
Separately, yes, you do normally sear it afterwards. It's for flavor/appearence/texture reasons though, and a number of people recommend mildly chilling it (say, for 15min or so) first so that you don't overcook any of the interior from the sear. The pan should be much hotter than a lot of people expect.
The only advantage is the automatic unit conversion, and that introduces a second point where the summary can get it wrong. If the source gives an incorrect answer then the AI summary is just going to repeat it, if you’re lucky.
No it wasn't, most of the first page results have the temperature right there in the summary, many of them with both F and C, and unlike the AI response, there is much lower chance of hallucinated results.
So you've gained nothing
PS
Trying the same search with -ai gets you the full table with temperatures, unlike with the AI summary where you have to click to get more details, so the new AI summary is strictly worse
Honestly the SEO talk sounds like reflexive coping in this discourse. I get that WWW has cheapened quality, but we now have the tech that could defeat most of the SEO and other trash tactics on the search engine side. Text analysis as a task is cracked open. Google and such could detect dark patterns with LLMs, or even just deep learning. This would probably be more reliable than answering factual queries.
The problem is there is no money and fame in using it that way, or at least so people think in the current moment. But we could return to enforcing some sort of clear, pro-reader writing and bury the 2010s-2020s SEO garbage on page 30.
Not the mention that the LLMs randomly lie to you with less secondary hints at trustworthiness (author, website, other articles, design etc.) than you get in any other medium. And the sustainability side of incentivizing people to publish anything. I really see the devil of convenience as the only argument for the LLM summaries here.
> But we could return to enforcing some sort of clear, pro-reader writing and bury the 2010s-2020s SEO garbage on page 30.
We could.
But it will absolutely not happen unless and until it can be more profitable than Google's current model.
What's your plan?
> Not the mention that the LLMs randomly lie to you with less secondary hints at trustworthiness (author, website, other articles, design etc.) than you get in any other medium. And the sustainability side of incentivizing people to publish anything. I really see the devil of convenience as the only argument for the LLM summaries here.
Well, yes. That's the problem. Why rely on the same random liars as taste-makers?
Why do you think that answer is correct? I mean maybe it is, or maybe it’s by the same user who recommended eating rocks (which ‘AI’ also recommended).
It doesn't take long to find SEO slop trying to sell you something:
When our grandmothers and grandfathers were growing up, there was a real threat to their health that we don’t face anymore. No, I’m not talking about the lack of antibiotics, nor the scarcity of nutritious food. It was trichinosis, a parasitic disease that used to be caught from undercooked pork.
The legitimate worry of trichinosis led their mothers to cook their pork until it was very well done. They learned to cook it that way and passed that cooking knowledge down to their offspring, and so on down to us. The result? We’ve all eaten a lot of too-dry, overcooked pork.
But hark! The danger is, for the most part, past, and we can all enjoy our pork as the succulent meat it was always intended to be. With proper temperature control, we can have better pork than our ancestors ever dreamed of. Here, we’ll look at a more nuanced way of thinking about pork temperatures than you’ve likely encountered before."
Sorry, what temperature was it again?
Luckily there's the National Pork Board which has bought its way to the top, just below the AI overview. So this time around I won't die from undercooked pork at least.
The link quoted does not have that text so what are you on about?
However that site gives the temperature for Pork as 71C which is not what USDA says but is correct. So using the USDA recommendation does have a risk according to at least Canada and UK
That's the thing, though — there isn't an objective standard here; it's mediated both by the local context (how good are the local trich inspections, etc.) and risk tolerance vs. cultural expectations for how the meat should taste. The Canadian and US governments currently disagree; so it goes.
Everything "has a risk". Taste and smell are not reliable indicators of bacterial contamination, and properly cooking meat won't eliminate dangerous toxins left behind by prior contamination if the meat was improperly stored before cooking.
Incredible, you are the problem. Didn't think I'd see such an idiotic answer on HN, please for the love of god do not use AI to know what is safe to eat.
id consider that google thinks its good enough for people to base their food safety off of it, and they deserve to get sued for whatever theyre worth for providing said recommendations when somebody trusts them and gets sick
How is it not the responsibility of senior management at a major retailer to ensure an exploit at a vendor can't take the whole house of cards down?
Many other major enterprise clients out there are all over vendor security/compliance ... auditing and reauditing vendors to minimise chance of this happening or worst-case, if does happen, containing it and recoverying quickly
>How is it not the responsibility of senior management at a major retailer to ensure an exploit at a vendor can't take the whole house of cards down?
I think you may be misunderstanding their organisation layout - his job is entirely to do with the quality of the products that they offer (and he's very good at it). He's nothing to with sales or online or any of that, but part of the 'normal' retail chain that people would never think goes anywhere this stuff. But their systems were all taken out because of this.
Last week stripped out all CSS from a fairly substantial project and replaced with Tailwind equivs, it got all but a few cases right
That was gemini-cli, I could see some mistakes on trial run so created a GEMINI.md with system prompt and project description (about 50 lines) which clarified some tricky source layout situations
Second run it was fine, ran for about an hour or so -- I had attempted to do it manually a while back but it started to look like it would take a week or two
Thanks for the insight. I have seen similar uses at work, where people do a bit of an enhanced codemod to migrate code from using one deprecated thing (library, function, syntax) to another. And while a codemod has to be more exactly programmed. AI gives you the ability to cover spots in the code that may not 1 to 1 fit with what the pattern you had in mind.
For the trivial cases that's fine (just using LLM does same)
But this particular project is not like a standard site and the CSS is in small fragments across 100s files and uses constants for some things like color values in places too
In that Loopple example you can see the conversion uses the Tailwind arbitrary value notation, the -[], so background-color:#afa8af gets converted to bg-[#afa8af], but I wanted nearest pure tailwind class bg-zinc-400, the agent seems to work out color distance fine so does all that in one-shot too
That's good to know it is better at translating the code from using one style to another! It is one of the gold use cases for AI agent coding at the moment. I've seen that at work as well.
I expect in a few days there will be a new tool launched that returns word frequency/velocity in recent biomedical papers ... so next year's PhDs can level things using an MCP function
Stack overflow didn't create a positive feedback loop where the solution to having to deal with an obscure, badly written, incomprehensible code base is creating an more incomprehensible sloppy code to glue it all together.
Neither did intellisense. If anything, it encouraged structuring your code better so that intellisense would be useful.
Intellisense does little for spaghetti code. And it was my #1 motivation to document the code in a uniform way, too.
The most important impact of tools is that they change the way we think and see the world, and this shapes the world we create with these tools.
When you hold a hammer, everything is a nail, as the saying goes.
And when you hold a gun, you're no longer a mere human; you're a gunman. And the solution space for all sorts of problems starts looking very differently.
The AI debate is not dissimilar to the gun debate.
Yes, both guns and the AI are powerful tools that we have to deal with now that they've been invented. And people wielding these tools have an upper hand over those who don't.
The point that people make in both debates that tends to get ignored by the proponents of these tools is that excessive use of the tools is exacerbating the very problem these tools are ostensibly solving.
Giving guns to all schoolchildren won't solve the problem of high school shootings — it will undeniably make it worse.
And giving the AI to all software developers won't solve the problem of bad, broken code that negatively impacts people who interact with it (as either users or developers).
Finally, a note. Both the gun technology and the AI have been continuously improved since their invention. The progress is undeniable.
Anyone who is thinking about guns in 1850 terms is making a mistake; the Maxim was a game changer. And we're not living in ChatGPT 2.0 times either.
But with all the progress made, the solution space that either tool created hasn't been changing in nature. A problem that wasn't solveable with a flintlock musket or several remains intractable for an AK-74 or an M16.
Improvements in either tech certainly did change the scale at which the tools were applied to resolve all sorts of problems.
And the first half of the 20th century, to this day, provides most of the most brilliant, masterful examples of using guns at scale.
What is also true is that the problems never went away. Nor did better guns made the lives of the common soldier any better.
The work of people like nurse Nightingale did.
And most of that work was that the solution to increasingly devastating battlefield casualties and dropping battlefield effectiveness wasn't giving every soldier a Maxim gun — it was better hygiene and living conditions. Washing hands.
The Maxim gun was a game changer, but it wasn't a solution.
The solution was getting out of the game with stupid prizes (like dying of cholera or typhoid fever). And it was an organizational issue, not a technological one.
* * * * *
To end on a good note, an observation for the AI doomers.
Genocides have predated the guns by millenia, and more people have died by the machete and the bayonet than by any other weapon even in the 20th century. Perhaps the 21st too.
Add disease and famine, and death by gun are a drop in the bucket.
Guns aren't a solution to violence, but they're not, in themselves, a cause of it on a large enough scale.
Mass production of guns made it possible to turn everyone into a soldier (and a target), but the absolute majority of people today have never seen war.
Much of the criticism of AI on HN feels driven by devs who have not fully ingested what is going with MCP, tools etc. right now as not looked deeper than making API calls to an LLM
The huge gap between the people who claim "It helps me some/most of the time" and the other people who claim "I've tried everything and it's all bad" is really interesting to me.
Is it a problem of knowledge? Is it a problem of hype that makes people over-estimate their productivity? Is it a problem of UX, where it's hard to figure out how to use these tools correctly? Is it a problem of the user's skills, where low-skilled developers see lots of value but high-skilled developers see no value, or even negative value sometimes?
The experiences seem so different, that I'm having a hard time wrapping my mind around it. I find LLMs useful in some particular instances, but not all of them, and I don't see them as the second coming of Jesus. But then I keep seeing people saying they've tried all the tools, and all the approaches, and they understand prompting, yet they cannot get any value whatsoever from the tools.
This is maybe a bit out there, but would anyone (including parent) be up for sending me a screen recording of exactly what you're doing, if you're one of the people that get no value whatsoever from using LLMs? Or maybe even a video call sharing your screen?
I'm not working in the space, have no products or services to sell, only curious is why this vast gap seemingly exists, and my only motive would be to understand if I'm the one who is missing something, or there are more effective ways to help people understand how they can use LLMs and what they can use them for.
My email is on my profile if anyone is up for it. Invitation open for anyone struggling to get any useful responses from LLMs.
> The experiences seem so different, that I'm having a hard time wrapping my mind around it.
Because we only see very disjointed descriptions, with no attempt to quantify what we're talking about.
For every description of how LLMs work or don't work we know only some, but not all of the following:
- Do we know which projects people work on? No
- Do we know which codebases (greenfield, mature, proprietary etc.) people work on? No
- Do we know the level of expertise the people have? Is the expertise in the same domain, codebase, language that they apply LLMs to?
- How much additional work did they have reviewing, fixing, deploying, finishing etc.?
Even if you have one person describing all of the above, you will not be able to compare their experience to anyone else's because you have no idea what others answer for any of those bullet points.
And that's before we get into how all these systems and agents are completely non-deterministic, and works now may not work even 1 minute from now for the exact same problem.
And that's before we ask the question of how a senior engineer's experience with a greenfield project in React with one agent and model can even be compared to a bon-coding designer in a closed-source proprietary codebase in OCaml with a different agent and model (or even the same, because of non-determinism).
> And that's before we get into how all these systems and agents are completely non-deterministic,
And that is the main issue. For some the value is reproducible results, for others, as long as they got a good result, it's fine.
It's like coin tossing. You may want tail all the time, because that's your chosen bet. You may prefer tail, but don't mind losing money if it's head. You may not interested in either, but you're doing the tossing and wants to know the techniques that works best for getting tail. Or you're just trying and if it's tail, your reaction is only "That's interesting".
The coin itself does not matter and the tossing is just an action. The output is what get judged. And the judgment will vary based on the person doing it.
So software engineering used to be the pursuit of tail of the time (by putting the coin on the ground, not tossing it). Then LLMs users say it's fine to toss the coin, because you'll get tail eventually. And companies are now pursuing the best coin tossing techniques to get tail. And for some, when the coin tossing gives tail, they only say "that's a nice toss".
> And companies are now pursuing the best coin tossing techniques to get tail.
With the only difference that the techniques for throwing coins can be verified by comparing the results of the tosses. More generally it's known as forcing https://en.wikipedia.org/wiki/Forcing_(magic)
What we have instead is companies (and people) saying they have perfected the toss not just for a specific coin, but for any objects in general. When it's very hard to prove that it's true even for a single coin :)
I think it's going to be personal. Because people define values in different ways, and the definition depends on the current context. I've used LLMs for things like shellscript, plotting with pyplot, explanations,... But always taking the output with a huge grain of salt. What I'm looking for is not the output itself, but the direction it can give me. But the only value is when I'm pressed for time and can't use a more objective and complete approach.
When you read the manual page for a program, or the documentation for a library, the things described always (99.99999...%) exist. So I can take it as objective truth. The description may be lacking, so I don't have a complete picture, but it's not pure fantasy. And if it turns out that it is, the solution is to drop it and turn back.
So when I act upon it, and the result comes back, I question my approach, not the information. And often I find the flaw quickly. It's slower initially, but the final result is something I have good confidence in.
> And often I find the flaw quickly. It's slower initially, but the final result is something I have good confidence in.
I guess what I'm looking for are people who don't have that experience, because you seem to be getting some value out of using LLMs at least, if I understand you correctly?
There are others out there who have tried the same approach, and countless of other approaches (self-declared at least) yet get 0 value from them, or negative value. These are the people I'm curious about :)
OP's comment also seems to be firmly stuck in 2023 when you'd prompt ChatGPT or whatever. The fact that LLMs today, when strapped into an agentic harness, can do or help with all of these things (ideation, architecture, use linters, validate code, evaluate outputs, and a million other things) seems to elude them.
Dothey do requirement gatherings? Like talking to stakeholder and getting their input of what the feature should, translating business jargon to domain terms?
No.
Do they do the analysis? Removing specs that conflict with each other, validating what's possible in the technical domain and in the business domain?
No.
Do they help with design? Helping coming up with the changes that impact the current software the least, fitting in the current architecture and be maintainable in the feature.
All they do is pattern matching on your prompt and the weights they have. Not a true debate or weighing options based on the organization context.
Do they help with coding?
A lot if you're already experienced with the codebase and the domain. But that's the easiest part of the job.
Do they help with testing? Coming up with tests plan, writing test code, running them, analysing the output of the various tools and producing a cohesive report of the defects?
I don't know as I haven't seen any demo on that front.
Do they help with maintenance? Taking the same software and making changes to keep it churning on new platforms, through dependencies updates and bug fixes?
Why do you think any of these should be a challenge for, say, O3/O3 pro?
You pretty much just have to ask and give them access for these things. Talking to a stakeholder and translating jargon and domain terms? Trivial. They can churn through specs and find issues, none of that seems particularly odd to ask of a decent LLM.
> Do they help with testing? Coming up with tests plan, writing test code, running them, analysing the output of the various tools and producing a cohesive report of the defects?
This is pretty standard in agentic coding setups. They'll fix up broken tests, and fix up code when it doesn't pass the test. They can add debug statements & run to find issues, break down code to minimal examples to see what works and then build back up from there.
> Do they help with maintenance? Taking the same software and making changes to keep it churning on new platforms, through dependencies updates and bug fixes?
Yes - dependency updates is probably the easiest. Have it read the changelogs, new api docs and look at failing tests, iterate to have it pass.
These things are progressing surprisingly quickly so if your experience of them is from 2024 then it's quite out of date.
> Do they do requirement gatherings? Like talking to stakeholder and getting their input of what the feature should, translating business jargon to domain terms?
No.
Why not? This is a translation problem so right up its alley.
Give it tool access to communicate directly with stakeholders (via email or chat) and put it in a loop to work with them until the goal is reached (stakeholders are happy). Same as a human would do.
And of course it will still need some steering by a "manager" to make sure it's building the right things.
> Why not? This is a translation problem so right up its alley.
Translating a sign can be done with a dictionary. Translating a document is often a huge amount of work due to cultural difference, so you can not make a literal translation of sentences. And sometimes terms don't map to each other. That's when you start to use metaphors (and footnotes).
Even in the same organization, the same term can mean different things. As humans we don't mind when terms have several definitions and the correct one is contextual. But software is always context free. Meaning everything is fixed at its inception and the variables govern flow, not the instruction themselves ("eval" instruction (data as code) is dangerous for a reason).
So the whole process is going from something ambiguous and context dependent, to something that isn't. And we do this by eliminating incorrect definitions. Tell me how LLMs is going to help with that when it has no sense of what correct and what it is not (aka judging truthness).
No they don't do requirements gathering, they also don't cook my food and wash my clothing. Some things are out of scope for an LLM.
Yes, they can do analysis, identify conflicting specs, etc. especially with a skilled human in the loop
Yes, they help with design, though this works best if the operator has sufficient knowledge.
The LLM can help significantly by walking through the code base, explaining parts of it in variable depth.
Yes, agentic LLMs can easily write tests, run them, validate the output (again, best used with an experienced operator so that anti-patterns are spotted early).
From your posts I gather you have not yet worked with a strong LLM in an agentic harness, which you can think of as almost a general purpose automation solution that can either handle, or heavily support most if not all of your points that you have mentioned.
I mean, if you "program" (prompt) them to do those stuff, then yeah, they'll do that. But you have to consider the task just like if you handed it over to a person with absolutely zero previous context, and explain what you need from the "requirements gathering", and how it should handle that.
None of the LLMs handle any of those things by themselves, because that's not what they're designed for. They're programmable things that output text, that you can then program to perform those tasks, but only if you can figure out exactly how a human would handle it, and you codify all the things we humans can figure out by ourselves.
> But you have to consider the task just like if you handed it over to a person with absolutely zero previous context,
Which no one does. Even when hiring someone, there's the basic premise that they know how they should do the job (interns are there to learn, not to do). And then they are trained for the particular business context, with a good incentive to learn well and then do the job well.
You don't just suddenly wake up and find yourself at an unknown company being asked to code something for a jira task. And if you do find yourself in such situation, the obvious thing is to figure what's going on, not "Sure, I'll do it".
I don't understand the argument, I haven't said humans act like that, what I said is how you have to treat LLMs if you want to use it for things like that.
If you're somehow under the belief that LLMs will (or should) magically replace a person, I think you've built the wrong understanding of what LLMs are and what they can do.
I interact with tools and with people. When with people, there's a shared understanding of the goal and the context (aka, alignment as some people like to called it). With tools, there's no such context needed. Instead I need reproducible results and clear output. And if it's something that I can automate, that it will follow my instructions closely.
LLMs are obviously tools, but their parameters space is so huge that's it's difficult to provide enough to ensure reliable results. With prompting, we have unreliable answers, but with agents, you have actions being made upon those reliable answers. We had that before with people copying and pasting from LLMs output, but now the same action is being automated. And then there's the feedback loop, where the agent is taking input from the same thing it has altered (often wrongly).
So it goes like this: Ambiguous query -> unrealiable information -> agents acting -> unreliable result -> unreliable validation -> final review (which are often skipped). And then the loop.
While with normal tools: Ambiguous requirement -> detailed specs -> formal code -> validation -> report of divergence -> review (which can be skipped) . There are issues in the process (which give us bugs) but we can pinpoint where we did wrong and fix the issue.
I'm sorry, I'm very lost here, are you responding to the wrong comment or something? Because I don't see how any of that is connected to the conversation from here on up?
>>> But you have to consider the task just like if you handed it over to a person with absolutely zero previous context, and explain what you need from the "requirements gathering", and how it should handle that
The most similar thing is software. Which is a list of instructions we give to a computer alongside the data that forms the context for this particular run. Then it goes to process that data and gives us a result. The basic premise is that these instructions need to be formal so that they became context-free. The whole context is the input to the code, and you can use the code whenever.
Natural language is context dependent. And the final result depends on the participants. So what you want is a shared understanding so that instructions are interpreted the same way by every participant. Someone (or the LLM) coming in with zero context is already a failure scenario. But even with the context baked in every participant, misunderstandings will occur.
So what you want is formal notation which removes ambiguity. It's not as flexible as natural language or as expressive, but it's very good at sharing instructions and information.
This is true, but they have helped prepare me with good questions to ask during those meetings!
> Do they do the analysis? Removing specs that conflict with each other, validating what's possible in the technical domain and in the business domain?
Yes, I have had LLMs point out missing information or conflicting information in the spec. See above about "good questions to ask stakeholders."
> Do they help with design? Helping coming up with the changes that impact the current software the least, fitting in the current architecture and be maintainable in the feature.
Yes.
I recently had a scenario where I had a refactoring task that I thought I should do, but didn’t really want to. It was cleaning up some error handling. This would involve a lot of changes to my codebase, nothing hard, but it would have taken me a while, and been very boring, and I’m trying to ship features, not polish off the perfect codebase, so I hadn’t done it, even though I still thought I should.
I was able to ask Claude “hey, how expensive would this refactoring be? how many methods would it change? What’s the before/after diffs on a simple affected place, and one of the more complex affected places look like?
Previously, I had to use my hard-won human intuition to make the call about implementing this or not. It’s very fuzzy. With Claude, I was able to very quickly quantify that fuzzy notion into something at least close to accurate: 260 method signatures. Before and after diffs look decent. And this kind of fairly mechanical transformation is something Claude can do much more quickly and just as accurately as I can. So I finally did it.
That I shipped the refactoring is one point. But the real point is that I was able to quickly focus my understanding of the problem, and make a better, more informed decision because of it. My gut was right. But now I knew it was right, without needing to actually try it out.
> Not a true debate or weighing options based on the organization context.
This context is your job to provide. They will take it into account when you provide it.
> Do they help with coding?
Yes.
> Do they help with testing? Coming up with tests plan, writing test code, running them, analysing the output of the various tools and producing a cohesive report of the defects?
Yes, absolutely.
> Do they help with maintenance? Taking the same software and making changes to keep it churning on new platforms, through dependencies updates and bug fixes?
Because MCPs solve the exact issue the whole post is about
reply