@gurupanguji / is on Nostr

Last Notes

2025-04-16 12:00:00

Neural networks > LLMs
https://youtube.com/playlist?list=PLZHQObOWTQDNU6R1_67000Dx_ZCJB-3pi&si=N9kF_jY15o1oDI7S
This is the playlist I always recommend to someone who’s trying to understand how does an LLM work.
#ai #llm #youtube

2024-12-22 04:19:56

ARC-AGI is a genuine AGI test but o3 cheated :( — LessWrong
ARC-AGI is a genuine AGI test but o3 cheated 🙁 — LessWrong:
They ask “how much of the performance is due to ARC-AGI data.” Probably most of it. If untuned o3 can do as well, don’t you think OpenAI would publish that (in addition to tuned o3)?
By training o3 on the public training set, the ARC-AGI no longer becomes an AGI test. It becomes yet another test of memorizing rules from its training data. This is still impressive, but something else.
I do not know what exactly OpenAI did. Did they let o3 spend a long time generating chains of thought, and reward ones which led to the correct answer? If that failed, did they have a human give examples of correct reasoning steps, and train it on that first? I don’t know.
They admitted they “cheated,” without saying how they “cheated.”
And there’s this in the end.
EDIT: People at OpenAI seem to be denying “fine-tuning” o3 for the ARC, see this comment by Zach Stein-Perlman. It’s unclear whether they’re denying reinforcement learning (rewarded by correct ARC training set answers), or whether they’re just denying they used a separate derivative of o3 (that’s fine-tuned for the test) to take the test.
It seems like o3 is definitely a step up. However, ARC being treated as a data set to solve ARC is definitely an argument that must be treated seriously.
#ai #llm #o3 #openai #reasoning

2024-11-13 01:14:04

The potential for a competitive market in technology build up and a bull case for TSMC
This is an interesting economic perspective on the technology industry today. Brad posits that AI build out is currently willing to pay an nvidia tax today
Once upon a time nearly everyone in tech paid the IBM mainframe tax. Then nearly everyone in tech paid the Microsoft-Intel DOS-Windows86 tax. Now nearly everybody pays the Google-Facebook-Amazon advertising tax, or the Apple iPhone AppStore tax, or both. And so everyone in tech today, perhaps especially the collectors of earlier taxes, desperately wants to avoid having to pay any future NVIDIA-OpenAI taxes for long. But they do so while everybody except Apple, incuding Google, is frantically paying an immense NVIDIA tax now.
Are the Tech Titans Mighty Enough to Break the Threatened NVIDIA-OpenAI Fetters?
This is despite efforts from every one of these infra build outs as well as frontier model generating companies also investing in their own efforts to move away from the Nvidia monopoly.
Pausing to build an alternative to NVIDIA and its CUDA, or to wait for a cheaper good-enough alternative to emerge, seems overwhelmingly risky in a tech world in which, in the past, too often first movers have become the last effective movers.
This is causing them to focus on the build out fueled by a fear to not be left behind. Now comes the interesting part – Brad posits that there are no “secret sauces” when it comes to any of the frontier models besides “time” and “compute”
This leads to a) why isn’t TSMC benefiting from the valuation that nvidia is able to get? Is CUDA that much better at harnessing the power of chip design from nvidia and b) when intel / samsung / qualcomm / some one else close the gap to nvidia’s design, there’s going to be some shifts to be seen.
#ai #chips #llm #mamlm