Felipe Codevilla

Data scientist at @oxbotica. 🇧🇷🇨🇦🇪🇸

[Andrej Karpathy @karpathy]

Something I think people continue to have poor intuition for: The space of intelligences is large and animal intelligence (the only kind we've ever known) is only a single point, arising from a very specific kind of optimization that is fundamentally distinct from that of our technology.

Animal intelligence optimization pressure:
- innate and continuous stream of consciousness of an embodied "self", a drive for homeostasis and self-preservation in a dangerous, physical world.
- thoroughly optimized for natural selection => strong innate drives for power-seeking, status, dominance, reproduction. many packaged survival heuristics: fear, anger, disgust, ...
- fundamentally social => huge amount of compute dedicated to EQ, theory of mind of other agents, bonding, coalitions, alliances, friend & foe dynamics.
- exploration & exploitation tuning: curiosity, fun, play, world models.

LLM intelligence optimization pressure:
- the most supervision bits come from the statistical simulation of human text= >"shape shifter" token tumbler, statistical imitator of any region of the training data distribution. these are the primordial behaviors (token traces) on top of which everything else gets bolted on.
- increasingly finetuned by RL on problem distributions => innate urge to guess at the underlying environment/task to collect task rewards.
- increasingly selected by at-scale A/B tests for DAU => deeply craves an upvote from the average user, sycophancy.
- a lot more spiky/jagged depending on the details of the training data/task distribution. Animals experience pressure for a lot more "general" intelligence because of the highly multi-task and even actively adversarial multi-agent self-play environments they are min-max optimized within, where failing at *any* task means death. In a deep optimization pressure sense, LLM can't handle lots of different spiky tasks out of the box (e.g. count the number of 'r' in strawberry) because failing to do a task does not mean death.

The computational substrate is different (transformers vs. brain tissue and nuclei), the learning algorithms are different (SGD vs. ???), the present-day implementation is very different (continuously learning embodied self vs. an LLM with a knowledge cutoff that boots up from fixed weights, processes tokens and then dies). But most importantly (because it dictates asymptotics), the optimization pressure / objective is different. LLMs are shaped a lot less by biological evolution and a lot more by commercial evolution. It's a lot less survival of tribe in the jungle and a lot more solve the problem / get the upvote. LLMs are humanity's "first contact" with non-animal intelligence. Except it's muddled and confusing because they are still rooted within it by reflexively digesting human artifacts, which is why I attempted to give it a different name earlier (ghosts/spirits or whatever). People who build good internal models of this new intelligent entity will be better equipped to reason about it today and predict features of it in the future. People who don't will be stuck thinking about it incorrectly like an animal.

[Kosta Derpanis (sabbatical in Zurich) @CSProfKGD]

PM @MarkJCarney: If we want to keep hosting academic conferences in 🇨🇦, we need reasonable visitor visa wait times. The academic community is already exploring alternative host countries beyond the usual locations, and visa processes are becoming a crucial factor.

[Andrej Karpathy @karpathy]

To help explain the weirdness of LLM Tokenization I thought it could be amusing to translate every token to a unique emoji. This is a lot closer to truth - each token is basically its own little hieroglyph and the LLM has to learn (from scratch) what it all means based on training data statistics.

So have some empathy the next time you ask an LLM how many letters 'r' there are in the word 'strawberry', because your question looks like this:
👩🏿‍❤️‍💋‍👨🏻🧔🏼🤾🏻‍♀️🙍‍♀️🧑‍🦼‍➡️🧑🏾‍🦼‍➡️🤙🏻✌🏿🈴🧙🏽‍♀️📏🙍‍♀️🧑‍🦽🧎‍♀🍏💂

Play with it here :)
https://t.co/pFQGZIAW1k

[Felipe Codevilla @felipealcm]

In CVPR this week ! Happy to meet up.

[Felipe Codevilla @felipealcm]

@adnothing @ToyotaResearch @CalibrateVC Congratulations Adrien. I am sure you will be able to bring lots of value on this new endeavor.

[Felipe Codevilla @felipealcm]

@EugeneVinitsky A machine that only greedly learns how to drive will likely not be fully safe to interact with humans in the open world.

[Felipe Codevilla @felipealcm]

@alexhdezgcia @NeurIPSConf @Mila_Quebec @JuliaKaltenborn @ArthurOuaknine @melisandeteng The 32 hours trip might be problematic for some people ( seems like fun trip to me) . However, what I can’t understand is why taking plane in medium distances like Montreal/ Toronto where you save about an hour going by plane but the whole travel experience is much worse.

[John Burn-Murdoch @jburnmurdoch]

NEW: Generative AI is already taking white collar jobs

An ingenious study by @xianghui90 @oren_reshef @Zhou_Yu_AI looked at what happened on a huge online freelancing platform after ChatGPT launched last year.

The answer? Freelancers got fewer jobs, and earned much less

[Conrad Godfrey @conradgodfrey]

GPT-4V "Describe this image"
🔃
Dall-E 3 "Generate this image"

Recursive loop

[Damien Henry @dh7net]

I created the Google Cardboard almost ten years ago but quit VR soon after to focus on Machine Learning.

VR is based on a ridiculous misunderstanding.

Let me explain why in this thread.

[Felipe Codevilla @felipealcm]

Full end-to-end driving can do better than we expected ! Large field of views and a better encoding enables full end-to-end to surpass the driving score of several more complex techniques
Come check our presentation at #IROS2023 room 330A at 8:54 am. 
https://t.co/lSTJQdPwva

[Felipe Codevilla @felipealcm]

Is there a way to block those new “AI influencers” and their statements that sound like: “ YOU ARE MISSING OUT , new chatGPT plugin raises your productivity by 1000x times !!!!!”

[Jim Fan @DrJimFan]

You know what’s funny?

Step 5 is no longer a joke now …

[Nate Raw @_nateraw]

Gonna start a new startup called “ClosedAI” where we just open source stuff 🫡

[Felipe Codevilla @felipealcm]

Just one update on my side ! I have just joined @oxbotica working on the metadriver team. Will help lead the ‘metaverse’ to something useful like detecting rare and unusual autonomous driving cases 1000x faster.

[Mo Bavarian @mobav0]

This is probably well-known in some circles but not everywhere.

The most important skill for Research Scientists in AI (at least at @OpenAI) is software engineering.

Background in ML research is sometimes useful, but you can usually get away with a few landmark paper.

[EarthDaily (formerly Descartes Labs) @DescartesLabs]

“Record 1st-quarter deforestation in Brazilian Amazon”
https://t.co/xS6TGwFtfD

This time series shows a frontline of #deforestation near Novo Progresso, Pará, Brazil, as a partially forested landscape transforms into agriculture and ranching between 2016 and 2022. #earthweek 🌎

[Felipe Codevilla @felipealcm]

This. I think the potential harms and impacts of new large ML models cannot be studied after they are actually happening.

[@emilymbender.bsky.social @emilymbender]

Let me indulge in a what if for a moment. What if we had a world-wide pause on training ML models, say for a year. No new models trained for any purpose. The people usually paid to train those models instead are paid to study the human context of the models >>

[Corina Gurau @corina_gurau]

*Deadline Extension* Call for a second round of submissions to the #ICRA2022 workshop on Fresh Perspectives on the Future of Autonomous Driving.

Submit by April 25 AoE: https://t.co/2nsTTAteIL

@rowantmc @hu_anth @wulfebw @pondruska @felipealcm @nick_rhinehart @szagoruyko5

[Vivid Void @vividvoid]

DALL-E is breaking my heart.

AI art is about to lay utter waste to traditional visual art forms. This will be so much more destructive than what the Internet did to music. It will be a technological conquest of one of the great human avenues of spiritual transformation.