The first thing I think and my friends think when trying to learn from AI is that it is fallible. ie it makes mistakes and the student has no way of knowing its wrong. GPT 3.5 made grammar mistakes when explaining simple Japanese sentences to me in English. even the CEO of Duolingo said that AI will have "small hits to quality" https://www.linkedin.com/posts/duolingo_below-is-an-all-hand... and my housemate said that the japanese he gets from ChatGPT 4o is awkward. (I told him to prompt it in Japanese instead of English and he said the result was a lot better)
Great points (and appreciate the coffeebreak recommendation!). Totally agree that the AI evaluation has plenty of inconsistency and errors.
I do want to clarify - it is more like audio only practice for digital flashcards. Meaning the prompt & response are both expected to be defined ahead of time. That way, GPT (as of today), is instructed to evaluate the semantic meaning of the user's response compared to the correct response.
they dont verify, they just present a LLM app and the user suffers if the information is not correct. however most of the time it is correct but sometimes it is not. one way to verify correctness is to ask a bigger model like OpenAI o3
I’m not sure how the solution to “LLMs lie” is “more LLM”. I’ve personally had o3 tell me stuff like: “I ran this query against version 0.10 of DuckDB and verified that it works” when the query contains functions that don’t exist in DuckDB or “this version gets better PageSpeed Insights results” when I know it can’t check those. It happens surprisingly often and it’s super obvious. However, it’s made me seriously wary of the information that it gives which I can’t verify purely based on logic.
well alot of action was required such as adding the model so no idea what happened to the guy who wrote the article maybe there is a new cursor update now
A website that extracts useful images from youtube videos. It would be a summary but includes the images because "text only" doesn't work for really new ideas. and
"text only" doesn't show bar graphs. And I think I will make it also make it draw boxes and arrows on top of the images for extra clarity.
>No one likes schleps, but hackers especially dislike them. Most hackers who start startups wish they could do it by just writing some clever software, putting it on a server somewhere, and watching the money roll in—without ever having to talk to users, or negotiate with other companies, or deal with other people's broken code. Maybe that's possible, but I haven't seen it.
>One of the many things we do at Y Combinator is teach hackers about the inevitability of schleps. No, you can't start a startup by just writing code. I remember going through this realization myself. There was a point in 1995 when I was still trying to convince myself I could start a company by just writing code. But I soon learned from experience that schleps are not merely inevitable, but pretty much what business consists of. A company is defined by the schleps it will undertake.
>The most striking example I know of schlep blindness is Stripe, or rather Stripe's idea. For over a decade, every hacker who'd ever had to process payments online knew how painful the experience was.
A lot of why people didn't build Stripe before was that to enter the payments space you needed connections to get the banks and payment processors to work with you. In comparison, you don't need anyone's permission to make uber for dry cleaners or something in line with other trends of the time. I doubt the Collison brothers would have been as successful getting Stripe off the ground if it had been their first company.
Working at a start-up now and seeing how many partnerships are solely due to connections of the CEO or a random board member is crushing. The tech side is an entirely and relatively easily solvable problem in comparison to the rest.
Has always been, there are a few cases where this wasn't true but odds are any industry you want to "enter" needs someone with connections to open the doors for you.
Tech is a problem that needs solving but it isn't the biggest problem to be solved, having a network and knowing people is more than half the job.
Those connections didn't materialize out of thin air, you know. Your CEO/random board member had to meet people, understand what they do, keep those connections alive, etc.
Why can't you do the same? Start now and in five years you'll be a lot more connected than you are at this moment. This is also an entirely solvable problem.
Banking is both deeply entrenched and well regulated. I suppose people could make a venmo/PayPal/cash app payment system but dealing with cards would be more difficult
Set a reminder for 10 years from now. Let's see how many incredible new tech products have been built. My guess is that ~any judge will decide that it turned out there were a lot of things still to be built
yeah. I worked on the internal banking connection at Square in the old days (~2011) and it was a _nightmare_. like, pre-TCP/IP connectivity that depended on dedicated copper to do teletype in COBOL-style fixed length fields in a cryptic format that was only specified in scans of paper documents. We had to write an adaptor that looked like, you know, REST on one end but then shoveled all of the traffic onto that single upstream connection, and then had to try to map responses that came back out of order to the right client. Miserable stuff and a threading nightmare.
This is hardly the point, but pg's use of schlep is jarring: it's primarily a verb ("to schlep"), but the noun form almost uniformly requires an article ("the schlep").
"No one likes schleps" should be "no one likes to schlep."
If AI coding assistants provide little value then why is Cursor IDE a 300m company and why does this study say it makes people more 37% more productive?
i like this advice. when trying to come up with new characters for fiction its very difficult to come up with something you don't already know but with this you have real people with their entire life story summarized for free.
Commercial Use License
If you wish to use a track from the mobygratis platform for commercial purposes, you must apply for a commercial license prior to such use. You may do so by sending an email to: mobygratis@moby.com and include your full name, the name of the Track, and your desired use of the Track, including all the commercial uses you anticipate, and the desired duration of the commercial license.
If one or more artists create an initial Collaboration and they are granted a commercial license, their revenue share would be, in the sole discretion of mobygratis, at most, 49% (forty-nine percent) of the gross income earned, received or credited from the permitted use and exploitation of that Collaboration. If a Collaboration is subsequently used by another artist or artists to create a new (or sub-) Collaboration—the new collaborating artist(s) will receive their share of any revenue exclusively from the initial collaborator(s); the mobygratis revenue share of any collaboration, regardless of how many layers of collaboration have occurred, shall always be greater than 50% (fifty percent). However, the specifics are subject to change in the sole and absolute discretion of mobygratis and would be covered on a case-by-case in the commercial license.
If you add > * text * it's more clear what is happening:
> Commercial Use License If you wish to use a track from the mobygratis platform for commercial purposes, you must apply for a commercial license prior to such use. You may do so by sending an email to: mobygratis@moby.com and include your full name, the name of the Track, and your desired use of the Track, including all the commercial uses you anticipate, and the desired duration of the commercial license.
> If one or more artists create an initial Collaboration and they are granted a commercial license, their revenue share would be, in the sole discretion of mobygratis, at most, 49% (forty-nine percent) of the gross income earned, received or credited from the permitted use and exploitation of that Collaboration. If a Collaboration is subsequently used by another artist or artists to create a new (or sub-) Collaboration—the new collaborating artist(s) will receive their share of any revenue exclusively from the initial collaborator(s); the mobygratis revenue share of any collaboration, regardless of how many layers of collaboration have occurred, shall always be greater than 50% (fifty percent). However, the specifics are subject to change in the sole and absolute discretion of mobygratis and would be covered on a case-by-case in the commercial license.
If trying to learn a language with only audio I recommend an AI-free audio podcast https://coffeebreaklanguages.com/coffeebreakspanish/
reply