deep learning and large language models. football (banter) fan. 19. prev: research intern at https://teapawt.org/t.co/Uyagq8HNqO
Mumbai, India.
Joined October 2019
- Tweets6.1K
- Following1.4K
- Followers977
- Likes34.2K
In Sept 2024, o1 surprised many purists who thought inference-time scaling for LLMs was through MCTS.
What if a connection exists, just implicit? What does it imply?
New post: "Squint enough and RLing CoT reasoners is approximable as Monte Carlo Tree Search policy learning." 🧵
to be clear: not ruling out human-led solutions completely, just that it's probably a surrogate problem to solve. which may just be a truism.
we'd pay the tax on the small part of the NN that needs to be editable (say engram), rest can remain a dense backbone with superposition and hard-to-interp features, no?
out with all CL talk talks about what we do to the learner, in with all the CL talk that talks about what the learner does to itself
To me it seems like if we are serious about this, we should focus on making arch choices that primarily optimise for "agentic gradient" as Will puts it. For ex: archs promoting cleanly separable reps may just be super easier for agents to brute-force on
teapawt.org/willcb/status/21124118…
feeling spicy: how continual learning gets solved is a hidden-in-sight race between when models get good enough to manually update themselves at weight or ctx level (direct edits on KV cache, engram modules, etc.) vs when humans find the correct recipe (unlikelier by the day)
Many trukes
My views on AI have radically shifted over the past few weeks, after spending way too much time reading about it and watching interviews of industry leaders and experts.
First, I really love the tools. I spend $200 on Claude every month and the product is worth every penny for me. Probably even more. I think the existing tools can deliver significant increase to future GDP and help us solve major global challenges faced by humanity.
However, I just cannot comprehend why the industry is not much more regulated, as its leaders and employees seem to request themselves. I meet a lot of people who have heard of the hugging face incident but don’t know the actual chain of events. If you don’t, you definitely should spend a least an hour reading about it.
I also do not know why we need to follow the path that frontier labs are following, to develop ever more powerful models. It seems to me that they are enormous gains that are readily available if we deploy the current technology in economic and scientific organizations without pursuing a super intelligence that they say might kill us and more likely wreak havoc in society. I know a lot of people think these warnings are marketing tricks, and that’s a reasonable assumption. But I don’t see the point of taking such risks.
My median scenario is that of many economists: AI will diffuse slowly and lift gdp growth by a percentage point or so, will cause some economic disruption on the way, but ultimately people will get new jobs and maybe work less.
However, I am deeply uncomfortable with the absolutely wild level of uncertainty around this prediction. And to be honest, even setting the doom scenarios aside, I find the utopia scenarios of some industry leaders actually dystopian. And I think a large majority of the public would agree if they paid attention to what they’re saying. To their credit, they themselves agree publicly that they should not be the ones making these choices. A large majority of the public is now very skeptical of AI as a source of human progress, in particular among younger adults who use it the most. They should have a say.
I don’t want to convince you of anything but to get informed about this topic, express yourself and vote accordingly.
General traits like tone, themes, voice are the one thing that reading the first few chapters (+ outros) do give you. It's literally just fingerprints (that can identify hands). I'm digressing but much of modern publishing literally runs on this, the intros etc. are all optimised by the author obsessively because this is true. How do you think publishers etc. sift thru stuff to decide what's worth publishing? How would you want ME to decide if investing a substantial amount of time to read that 145-page PDF is worth it without picking up tones, cues, etc.?
Hey you can say what you want and call it "tasteless". But you first said "less than 30 minutes", now you say 15. It was actually ~36. Just FWIW.
BTW, as I say in the main tweet, my gripes are mainly with tonality, general posturing and this general line of "thonkposts" (and not the forecasting which may or may not be correct, which is what a fuller read would've exposed me to). Though I own up for the fact that I made a mistake not prefacing the tweet with "just skimmed thru but...". That wasn't good!
Def not "critique automated by AI". It's what I sincerely felt after skimming thru parts of it (and reading summaries of latter ones).
Hiranmay Darshane retweeted
glimpses of the future
Aware is (basically) a browser we're building to explore personal computing and generative, malleable software. it keeps a footprint of the materials (pages, files) as a graph in a relational database stored locally.
Been writing my paper in my own latex editor inside Aware. It's very satisfying when software fixes itself and fetches its own context and attach them to the right parts.
The editor is an experiment of "interface as a skill": the editor's code and its skill live in the paper's repo, so the same agent that knows my code and data also lives in the editor. So i get:
1. an interface i can reprogram any part anytime (aesthetics i like, and the scratch pad i always wanted)
2. an agent living in the editor, with the whole project as context (a figure links to the dashboard, built from the real analysis code that made it)
(though i have to admit, some of my focused writing time now goes to wanting to tweak the editor instead)
It doesn't make sense to savor the doom like that post does. IMO it's better to say “I can't see a world where this goes well, and that scares me” instead of "Well hey this is how it must be, rejoice”.
teapawt.org/WillManidis/status/211…
we should suggest mitigations and interventions (including pauses if necessary). I'm mainly put off by the super prophesying language such posts use "doom is inevitably coming, etc." and sometimes even glorifying/embellishing it smugly.
FWIW I was referring to this post:
The benefits of abundant intelligence will accrue far outside of San Fransisco.
Spent the last year writing about capital, sovereignty, and the coming reordering of economic power.
THE COMMANDING HEIGHTS
Commanding-Heights.com
All the thonk-bois high on their own thought, writing forecasts (often super negative and dystopian ones), must remember, however logically plausible or agreeable those forecasts may be, that these are times when even writing should have empathy for the masses and a general texture of sobriety.
A doctor can’t say “hey, you have cancer” and just walk out of the room. Nobody’s asking the doctor to lie about the diagnosis. If they believe it’s true, say it, and say it plainly. Then stay in the room, put a hand on the shoulder, and talk about what can be done next.
Writing “displacement is coming, everyone will lose their jobs, but this must happen anyway, the line must go up, it’s some great fundamental law of nature” is like the doctor delivering the diagnosis, adding that treatment is pointless because it’s all the natural order of things, and then leaving. Dressing up inevitability as wisdom (thinking it's some sigma alpha thonk-boi hardass moment) is just corrosive and distasteful.
Be as bleak as the evidence warrants (or as much as you want to be). But these are times when one should write like the people most affected in your forecasts are reading it, because they probably are, and doom-prophecies just by themselves help very little (only reflect poorly on you).