npsomaratna

I remember synthesizing H2S in high school (for chemistry practicals, I think). I made the mistake of attempting to smell it. This was for a fraction of a second—but during that time, I could feel the individual bubbles of gas entering my nose; and I almost passed out.

I adopted Chrome almost as soon as it was released. Chrome was better than the other browsers available at that time. Firefox was a mess, and Internet Explorer... well, best left unsaid.

Quick question: what would you estimate the running cost of Llama 2 70b to be? (On GPU, and assuming maximum utilization)?

Agreed. Worse, you don't want to be in a position where any single company can end your business at the flip of a switch.

We have. This is acceptable for some clients, but not for others. Both groups, however, prefer maintaining complete control over their data, given the chance.

Edit: plus, my personal view is that local LLMs are the future. They've already caught up to GPT-3.5 (based on my testing); and they continue to evolve rapidly. Makes sense to focus our limited resources on riding that wave.

OpenAI won't go away, but neither will they remain the first choice (or only choice!) for most use-cases.

CRLF

Data security and privacy. Our clients (in aviation, finance, etc.) need this due to legal and regulatory reasons. Also, the new Llama 2 models are very powerful. In my testing, Llama 2 70b is comparable to GPT-3.5 in capability.

(Shameless plug: here's our website: https://www.amw.ai/)

If I remember correctly, Enrico Fermi’s wife was Jewish. This forced him to migrate to the U.S.

My understanding is that in NTK aware RoPE scaling, the model does pay uniform attention. With older methods, not as much.

There is a magic about the character movements in PoP. I recently replayed it—almost 30 years after I first played it. There is a fluidity in how the prince moves that is still impressive today.

I understand that Jordan Mechner put in a ton of hours rotoscoping the movements. (Example: https://www.youtube.com/watch?v=fCy8WqAgPbY)

On a tangent, Mechner is an incredibly talented (and in my opinion, severely underrated) game designer. "The Last Express" is a masterpiece, and wondrous in a different way.

Yes. Whether we like it or not, AI is with us to stay. A skill that AI can easily supplant is a skill that will become outdated very quickly. We're better off teaching students how to use AI effectively. Hopefully this will "future proof" them somewhat.

Same here. The researchers I know in medicine all seem to use word.

I think you'd need to offload the model into CoreML to do so, right? My understanding is that none of the current popular inference frameworks do this (not yet, at least).

Out of curiosity, what would you consider Finance 101?

Yup. In Sri Lanka we calls chilies "miris," and pepper "gam miris" (translation: village chilies).

I guess folks originally used pepper as the way to set their mouths on fire; and once introduced, happily jumped over to chilies as a much better way to do so.

Not an academic here, but I've read (and continue to read) through research papers regularly.

The original bitcoin paper is a great example. I was able to follow the paper almost fully at my first read itself—despite my not having a formal background in maths.

...and as you said, many of the insubstantial papers hide behind jargon and unnecessarily complex equations, just to camouflage their lack of substance. It's frustrating to spend time deciphering a paper, only to realize that you've essentially wasted that time.

My experience with RAG is that while it reduces the incidence of hallucinations* significantly (especially if you reduce the LLM temperature to zero at the same time), it doesn't eliminate them.

My startup has a product for lawyers that uses RAG to answer legal queries (https://lawlight.ai/). We have a disclaimer that "... (we) do not guarantee the accuracy of answers. You are responsible for reviewing the cited case law and drawing your own independent conclusions."

(This works within the specific context—lawyers are domain experts; and they are supposed to read through all cases they cite in court anyway.)

* I dislike the term "hallucinations." By definition LLMs hallucinate. It's just that much (or most) of the time, the hallucinations reflect reality.

Mainly for data processing, and almost exclusively "open-source" LLMs now (essential given our focus on industries with tons of confidential data). Let me plug my startup's website, which describes what we do: https://www.amw.ai/

This isn't a parody, right? I mean, surely no-one would (seriously) say this in public in real life?

Right?

Re: step 3, I think you mean text-embedding-ada-002—OpenAI's current embeddings model, which replaces all 15 (or is it 16?) first generation embeddings models.

With respect to open source embeddings models, instructor-xl is the state of the art currently—as effective as text-embedding-ada-002.

That said, instructor-xl has a context length of 512 tokens, while text-embedding-ada-002 has a context length of 8192 tokens, which is markedly more convenient.

Last but not least, parent's comment re: langchain is spot on. It's simple and straightforward to write these few lines of code yourself.

The stuff I do with LLMs is pretty conventional and wouldn't trigger even the most conservative content filter. However, the idea that a company can gatekeep LLMs to the point where you're at their mercy makes me feel angry.

I'm glad that Llama 2 and other open source (let's not fight about semantics please) LLMs are catching up. OpenAI and Microsoft need serious competition.

Going on a tangent here: there's a reason medicine is heavily regulated—because of the lessons we've learned the hard way. The same reason law is also heavily regulated. Some of these lessons have become obscured by time, unfortunately. Other lessons may not be fully applicable to the modern state of the field. (I say this as a qualified doctor and lawyer with some appreciation of the history of both fields.)

Are the current regulation regimes perfect? No. Would we be better off without it? I doubt it. Should we try to strike a balance between regulation and innovation? Yes—but doing so is devilishly tricky. No matter what you do, someone somewhere will be unhappy.

If someone from Anthropic is reading this thread: do consider me too ;) I've submitted 3 applications over a period of several months so far. Complete radio silence.

At this point, Anthropic completely ignoring waitlist applications (without even a confirmation) is almost a meme. By the time they open up, a lot of devs will be entrenched in OpenAI's arms.

This is unsubstantiated. The only folks who know exactly how GPT-4 works are employed at OpenAI. The rest of us can only guess.

That reminds me of the undocumented LOADALL instruction, and how you could use it to access memory without switching to protected mode. Good times!

Agreed. I don't really have time to play PC games anymore, but when I did, I fell into the demographic you described.

(Maybe not 4k 60fps—after all, one does have to compromise sometimes...)

Could someone more knowledgeable explain what the practical effect of this license change is? As far as I can see, there isn't much of a difference?

That's interesting. Do you mean that you ask GPT-4 to break down it's own prompt into smaller/simpler pieces?

I am surprised that this article (which appears to be a marketing piece with superficial content) ranks so high on HN (#3 at the time of writing).

There are multiple inaccuracies. In particular, the article implies that tokens are universal to all AI—which is incorrect. For example, the statement that "AI systems depend on visual tokens to make sense of images" is wrong. This might be true of vision transformers, but not of CNNs.

This reflects my experience from the 00's onwards. Survived and avoided most of these—except the nosql bandwagon. A worn out and wiser man, ended up switching back to relational DBs several years later.