ironrabbit

Which part of SB1047 do you think is too onerous or not "slow regulation" enough?

This is a forum post from a student with 18 total karma

Zero chance private github repos make it into openai training data, can you imagine the shitshow if GPT-4 started regurgitating your org's internal codebase?

Automatic kernel fusion (compilation) is a very active field, and most major frameworks support some easy-to-use compilation (e.g. jax's jit, or torch.compile which iirc uses openai's triton under the hood). Often you can still do better than the compiler by writing fused kernels yourself (either in cuda c++ or in something like triton (python which compiles down to cuda) but compilers are getting pretty good.

edit: not sure why op is getting downvotes, this is a very reasonable question imo; maybe the characterization of kernel compilation as "AI" vs. just "software"?

The median OAI employee with a 900k comp is probably L5, not L6 or L7

New hires' comp is much higher than existing employees', especially if you've hit your cliff. 7 figures for E6 can happen if you joined recently, have good counter-offers, and negotiate. It's not super uncommon but it's also not the median E6 comp.

All engineers and researchers, even junior, are "Member of Technical Staff" at OAI

> the company pushed back against a proposed amendment to the AI Act that would have classified generative AI systems such as ChatGPT and Dall-E as “high risk” if they generated text or imagery that could “falsely appear to a person to be human generated and authentic.” [...] The company argued that it would be sufficient to instead rely on another part of the Act, that mandates AI providers sufficiently label AI-generated content and be clear to users that they are interacting with an AI system.

This sounds pretty reasonable? I don't think it's hypocritical to be talking about the doom of humanity and also arguing that GPT-3, a 3-year old model, should not be classified as "high-risk" in that sense.

Even if you disagree, questioning Altman's leadership and calling him an "empty soul" over this kind of regulatory detail is not adding substance to the discussion imo.

Character has their own models, and anecdotally I've heard they have one of the better LM training codebases out there.

Openai is probably the highest status employer among my network, but I don't know any university students

Andrej's "Building Makemore" series is exactly this, it includes a wonderful lecture where he computes all the gradients for a simple network by hand and compares them against the values produced by torch's autograd.

minGPT prioritized being understandable above all else, and was not very fast. This repo includes several optimizations, but it still much more understandable than probably any other open source implementation.

Anecdotally, chatgpt seems much worse to me than Google for getting correct answers. Like orders of magnitude worse. Tells me the wrong timezone for a city kind of bad. No doubt it will be much better in the future, and they've definitely found PMF with the interface, but I would not trust it right now with anything even slightly important to me.

This looks very nice but no android support is unfortunately a dealbreaker