Will Rice

ML Engineer working on generative models in #Speech and #NLP. Focused on Text-to-Speech (TTS) and speaker generation.

[Will Rice @_Will_Rice]

@wandb

[Will Rice @_Will_Rice]

Is making a team and paying for a seat the only way to get notifications for @wandb anymore?

[Will Rice @_Will_Rice]

@karpathy Mostly agree, except I’ve totally had Claude code suggest to add ignore comments to mypy errors after tying to fix it. So I think it does get stuck and give up. The difference is you can just say “don’t reference my other chats” and it will try again.

[Will Rice @_Will_Rice]

@sedielem I put normalizing flow, but seeing flow-based in an abstract has changed over the years what I expect the paper to be about. Early in my career I would expect Real NVP, but now I would expect flow matching.

[Will Rice @_Will_Rice]

I find deep research from @OpenAI super useful. However, I would like to see improvement in the output format. I always want it to be in markdown but it leaves out information when asked to convert it to markdown. If anyone has luck with getting the full output in a markdown file, I would appreciate any tips.

[Will Rice @_Will_Rice]

what do you do when you get rate limited by your AI therapist?

[Will Rice @_Will_Rice]

I'd like to see the perfomance of a random sample from the same companies ChatGPT was given to pick from for a comparison.

[Rohan Paul @rohanpaul_ai]

Its going viral on Reddit.

Somebody let ChatGPT run a $100 live share portfolio, restricted to U.S. micro-cap stocks.

Did an LLM really bit the market?.

- 4 weeks +23.8%

while the Russell 2000 and biotech ETF XBI rose only ~3.9% and 3.5%.

Prompt + GitHub posted

---

ofcourse its a short‑term outperformance, tiny sample size, and also micro caps are hightly volatile.

So much more exahustive analysis is needed with lots or more info (like Sharpe ratios and longer back-testing etc), to explore whether an LLM can truly beat the market.

[Will Rice @_Will_Rice]

So we tuning on the test set now?

[Will Rice @_Will_Rice]

When is @astral_sh going to replace mypy? 😁

[William @williamwelsh]

@thdxr we heard you like abstracting so we abstracted your abstraction to give you more abstraction 🤩

[Will Rice @_Will_Rice]

Someone remind me how many years it was between the first GAN paper and realistic faces that still had artifacts.

[ROM 🥔 @i_dg23]

more detailed version

[Will Rice @_Will_Rice]

@timClicks If you look at the code for the paper there is a lot of triton cuda stuff

[Will Rice @_Will_Rice]

@JeremiahDJohns It’s interesting that some of these came from sources already on the internet that are obviously trolls/satire. That actually might be fixable if you could rank existing answers based on truthfulness.

[Will Rice @_Will_Rice]

Not surprising to anyone that’s actually worked on generative models.

[Riley Goodside @goodside]

GPT-3’s score on the MMLU benchmark was 40%. First release of GPT-4 scored 86%, and today GPT-4o is 89%.

An increase of just 3% — that’s a full year of progress. If you plot the prior trend we were supposed to be at 100, maybe 120% by now. AI is hitting a wall.

[Will Rice @_Will_Rice]

Unpopular opinion: AI agents are going to do some things well, but fail to generalize enough to be useful.

[Nick Davidov @Nick_Davidov]

Unpopular opinion: AI agents are hard.

[Will Rice @_Will_Rice]

@jankosinski How do you know that it was accurately predicted?

[Arun Rao @sudoraohacker]

Hot take: ML researchers are underestimating how quickly recent scaling laws may flatten out - it’s quite likely what people see as an exponential function is a sigmoid, and the harsh reality of the physics of high energy costs and power plant construction restrict the expected benefits from pretraining of ever larger models.

[Will Rice @_Will_Rice]

tbh esembling would be too slow for the notebook submission

[Will Rice @_Will_Rice]

Top 10 on #kaggle and I'm not even ensembling. https://t.co/BFQ1htWSZF

[Will Rice @_Will_Rice]

Moved up to rank 19 #kaggle https://t.co/BFQ1htWSZF