I don't think the narrative makes sense. It was clear from way back in 2016 that training would take a ton of resources. Researchers were already been sucked into FAANG labs because they had the data, the compute, and the money. There was never a viable way for a true non-profit to make world-changing, deep learning-based AI models.
When seen through the rearview mirror, the whole narrative screams of self-importance and duplicity. GPT-2 was too dangerous, and only they were trust-worthy enough to possess. They were trust-worthy because this was a non-profit, so "interest aligned with humanity". This charade has continued even to barely some months ago.
My bet: they use formal methods (like an interpreter running code to validate, or a proof checker) in a loop.
This would explain: a) their improvement being mostly on the "reasoning, math, code" categories and b) why they wouldn't want to show this (its not really a model, but an "agent").
I think it could be some of both. By giving access to the chain of thought one would able to see what the agent is correcting/adjusting for, allowing you to compile a library of vectors the agent is aware of and gaps which could be exploitable. Why expose the fact that you’re working to correct for a certain political bias and not another?
Jokes aside, carpentry is an amazing complementary hobby for software related work. There's something about drawing on a piece of paper and using your hands to make it a reality that scratches a lot of itches.
Sadly, entrepreneurship doesn't leave much free time for carpentry.
Getting into manual machining is another good one. If you go into it with even a hint of a "move fast and break stuff" mentality you'll find yourself mostly just breaking a lot of stuff. It's good practice for being deliberate in your actions.
Yes, I originally thouoof building AI workflows with an existing DAG package or creating my own.
But I think DAGs don't scale, or rather, as you point out, it quickly becomes easier to reason about code than about a graph.
I thought, maybe a setup like pytorch that let's you code it normally and once you run it or compile it creates a graph for you to see. But I remain unconvinced.
I like the Pytorch comparison, and I've seen DSPy position themselves as Pytorch for prompting.
I also think the actor model is a natural fit for AI agents, which has some similarities with Pregel (message passing), and some differences (there are no super-steps of graph execution, each actor has its own thread).
I definitely dislike state machines for most use cases.[2] I think we can learn a lot about good AI agent paradigms from game programming, and I enjoyed this article on game state: https://gameprogrammingpatterns.com/state.html
At the end, they mention that game AI doesn’t often use state machines anymore, because the structure they impose is limiting.
Also, the folks behind Temporal are anti-state-machine:
[1]: our goal is to be an open-source alternative to the OpenAI Assistants API, not to compete with LangGraph, but there is overlap
[2]: I understand that LangGraph is not a state machine
Yea, I think it clicked with me when I saw the DSPy documentation that maybe that's a good balance between code-like and graph like, though I still find DSPy to be overly obfuscated.
I just use magicplan (a freemium app) to create floor plans just by scanning with my phone. Of course it won't be as accurate as a proper 3D scanner but prospective renters aren't going to notice. You can of course edit the floor plan afterwards.
We bring human-like decision making into game mechanics.
We do this via behavior systems that rely on a combination of generative AI and planning algorithms. The most obvious use case is AI NPCs, but game masters, storytellers, opponents, and a myriad of other game mechanics can be greatly enhanced by the same core technology.
We are looking for a game developer who has deep experience with game design and is a proficient programmer. Hence the title.
Your role would primarily be to build experimental games that use the tech.
To apply, just send me an email at ramon@clementine.games with the title "HN Hiring - Technical Designer".
About us:
Well, me, for now. I am Ramon, founder, and I am working from San Francisco. You can find more about me in my personal website (http://ramondario.com/). I will be joined by an AI engineer in September.
I got some funding from A16Z Speedrun and am assembling a team of AI/ML engs and game devs.
p.s. If you find this interesting, but are not a game developer, feel free to contact me regardless!
When seen through the rearview mirror, the whole narrative screams of self-importance and duplicity. GPT-2 was too dangerous, and only they were trust-worthy enough to possess. They were trust-worthy because this was a non-profit, so "interest aligned with humanity". This charade has continued even to barely some months ago.
reply