I disagree with the premises outright. I have no intention of causing no harm to anyone. Just by living, we cause harm to others and other beings.
I am going to say things to other people and that is going to make them upset, that is ok. They are going to say things that make me upset, that is ok.
Someone is going to give me a cold, flu, etc. and that is ok. They didn't have to do it, they could have stayed inside for weeks until they were sure they were not sick, but who wants to live that way?
There is a level of harm in anything that that a living thing does.
It's not a question of 'stay inside for weeks' or 'go out' though is it? It's a spectrum and there are a lot of basic precautions like masking possible that people refuse to do.
Personally I live my life but will continue wearing a KN95 until Covid stops being in active circulation. You can call this paranoia but I haven't caught anything since I started doing this and taking other basic precautions like distancing - no covid, no flu, no colds. So I'm benefiting from it and by doing this I'm less likely to spread a disease to someone immunocompromised.
Earlier during covid I saw a lot of people doing the same, which I appreciated. Now I'm the only person doing this and it makes me a little sad, because if I spread a cold or flu or covid to somebody's grandma, they might die even if the virus is no problem for my younger immune system.
In other countries it's conventional to wear a mask any time you're remotely sick or have symptoms that might indicate illness, but we don't do it here. Instead, people go to work while sick and make their coworkers sick.
the point was not to get into a mask debate but "In other countries it's conventional to wear a mask any time you're remotely sick" should ring as BS to you, because those countries have flu and cold just like any other country.
AND if you ask them, they don't do it to prevent others from getting sick (not that that works unless you are gasping air through an n95), but they do it to protect themselves. This is especially true in China where there is also air pollution to consider.
I wasn't being literal when I said 'who wants to live that way' as in to mean every single person with no exceptions. yes, Locked-in syndrome is a thing. its not healthy but hey, you do you.
I can exercise at home, I moved outside the city and I have a large patio so I get sun and fresh air, I just only leave the house less than once a week when there's something I can't do from home. My friends all moved to different countries years ago so I only see them online anyway. I do yearly check ups, I'm actually in much better health than when I lived in the city and went outside several times a day.
Yes. But in this case it's not known who did this. One NATO member is trying to pin it on Ukraine. But evidence is scarce. Personally I'm 50/50 whether this was russian false-flag or combined effort of some NATO members and Ukraine.
So long as we live in democracies, we are responsible for the actions of our governments.
You can certainly go "Don't blame me, I voted for Kodos" during domestic discussions of displeasure of the ruling party. But, in international affairs, we are accountable for our government's foreign policies.
I'm Canadian. "We" are in a proxy war with Russia. "We" need to win lest Putin thinks he can just take sovereign nations like Ukraine without the rest of the world stopping him.
Appeasing dictators is a losing policy. "We" need to do everything possible by having Europe fund Ukraine, and now while "We" have Biden agreeing to do so, until Trump takes over and "we" have infighting between NATO nations about what to do about Ukraine.
I have found that embeddings + LLM is very successful. I'm going to make the words up as to not yield my work publicly, but I had to classify something into 3 categories. I asked a simple llm to label it, it was 95% accurate. taking the min distance from the word embeddings to the mean category embeddings was about 96%. When I gave gave the LLM the embedding prediction, the LLM was 98% accurate.
There were issues an embedding model might not do well on where as the LLM could handle. for example: These were camel case words, like WoodPecker, AquafinaBottle, and WoodStock (I changed the words to not reveal private data).
WoodPecker and WoodStock would end up with close embedding values because the word Wood dominated the embedding values, but these were supposed to go into 2 different categories.
> word Wood dominated the embedding values, but these were supposed to go into 2 different categories
When faced with a similar challenge we developed a custom tokenizer, pretrained BERT base model[0], and finally a SPLADE-esque sparse embedding model[1] on top of that.
Do you mind sharing why you chose SPLADE-esque sparse embeddings?
I have been working on embeddings for a while.
For different reasons I have recently become very interested in learned sparse embeddings. So I am curious what led you to choose them for your application, and why?
> Do you mind sharing why you chose SPLADE-esque sparse embeddings?
I can provide what I can provide publicly. The first thing we ever do is develop benchmarks given the uniqueness of the nuclear energy space and our application. In this case it's FermiBench[0].
When working with operating nuclear power plants there are some fairly unique challenges:
1. Document collections tend to be in the billions of pages. When you have regulatory requirements to extensively document EVERYTHING and plants that have been operating for several decades you end up with a lot of data...
2. There are very strict security requirements - generally speaking everything is on-prem and hard air-gapped. We don't have the luxury of cloud elasticity. Sparse embeddings are very efficient especially in terms of RAM and storage. Especially important when factoring in budgetary requirements. We're already dropping in eight H100s (minimum) so it starts to creep up fast...
3. Existing document/record management systems in the nuclear space are keyword search based if they have search at all. This has led to substantial user conditioning - they're not exactly used to what we'd call "semantic search". Sparse embeddings in combination with other techniques bridge that well.
4. Interpretability. It's nice to be able to peek at the embedding and be able to get something out of it at a glance.
So it's basically a combination of efficiency, performance, and meeting users where they are. Our Fermi model series is still v1 but we've found performance (in every sense of the word) to be very good based on benchmarking and initial user testing.
I should also add that some aspects of this (like pretrained BERT) are fairly compute-intense to train. Fortunately we work with the Department of Energy Oak Ridge National Laboratory and developed all of this on Frontier[1] (for free).
actually, that's what makes chat gpt powerful. I like an LLM willing to go along with what ever I am trying to do, because one day I might be coding, and another day I might be just trying to role play, write a book, what ever.
I really cant understand what you were expecting, a tool works with how you use it, if you smack a hammer into your face, don't complain about a bloody nose. maybe dont do like that?
It's not good for any entity to role play without signaling that they are role-playing. If your premise is wrong, would you rather be corrected, or have the person you're talking to always play along? Humans have a lot of non-verbal cues to convey that you shouldn't take what they are saying at face value - those who deadpan are known as compulsive liars. Just below in them in awfulness are people who don't admit to having being wrong ("Haha, I was just joking" /"Just kidding!"). The LLM you describe falls somewhere in between, but worse if it never communicates when it's "serious" and when it's not, and bot even bothering with expressing retroactive facetiousness.
I didn't ask to roleplay, in this case it's just heavily hallucinating. If the model is wrong, it doesn't mean it's role-playing. In fact, 3.5 Sonnet responded correctly, and that's what's expected, there's not much defense for GPT-4o here.
So if you're trying to write code and mistakenly ask it how to use a nonexistent API, you'd rather it give you garbage rather than explaining your mistake and helping you fix it? After all, you're clearly just roleplaying, right?
I found the most anomalous response was as good (15/20) or better (5/20) than the temperature 0 response in 20 samples.