Browning.jake00 @Jake_Browning00 Twitter profile

Last Seen Profiles

@jandakembangstw

@bookawardpro

@jerkimof

@Ella_Addison768

@loveearthtone

@mimo00957910011

@iBankIntellect

@olicelvr

@sebavallejo

@jackmogs69

@ruroken_movie

@shytaylax

@Yoru25072002

@HNYCancer

@lirpaaseever

@becky_storrie

@IndieBookButler

@penny4ya

@LatinaMaha71073

@BBwriterchick

@michael_galkin

@HotMikeCommish

@DenizRtprensesi

@stw_pdg

@s_m_273

@mhnd78413106

@jingai2828_moe

@GoodnitesGamer

@LaceyBlair1155

@SudanTribune_AR

@aironmaso

@OficialHinava

@Maltgames

@TorraMahno

@J0CwdGKQKTjt11

@RandyPete

Browning.jake00

@Jake_Browning00

9 months

Now that both Altman and Gates are acknowledging scaling up won't help these models improve, I think it is safe to pronounce a verdict on the current approach to generative AI: it's boring.

37

62

405

Browning.jake00

@Jake_Browning00

1 year

I don't think anyone should report on the 22-word letter from Altman et al. on regulating AI to prevent existential risk without mentioning it comes just after OpenAI threatened to leave the EU rather than be regulated.

Focus: OpenAI CEO's threat to quit EU draws lawmaker backlash

For months, Sam Altman, CEO of Microsoft-backed OpenAI has urged lawmakers around the world to draw up new rules governing the technology. On Wednesday, he threatened the ChatGPT maker may leave the...

www.reuters.com

15

63

190

Browning.jake00

@Jake_Browning00

2 years

A piece by @ylecun about whether deep learning is hitting a wall--or just facing another hurdle. A response to @GaryMarcus attempting to lay out the different perspectives and why the debates get so heated.

What AI Can Tell Us About Intelligence | NOEMA

Can deep learning systems learn to manipulate symbols? The answers might change our understanding of how intelligence works and what makes humans unique.

www.noemamag.com

11

37

172

Browning.jake00

@Jake_Browning00

10 months

I don't know if IIT is pseudoscience, but it is pernicious. While feigning scientific respectability, it avoided rigor and peer review in favor of pop books that obscured or ignored decades of consciousness research. The field will be better without it.

"Consciousness Winter" wasn't that cold--and the "spring" generated more heat than light

There is a story that goes, "in the 20th century, you couldn't talk about consciousness. Before Edelman, Crick, and Koch, it was a bad word that no one would touch." The recent criticism of the...

www.jacob-browning.com

9

17

93

Browning.jake00

@Jake_Browning00

9 months

Written texts aren't the same as human knowledge & pretrained LLMs aren't reading those texts. They're trying to reproduce the words, not figure out how to apply them. LLMs aren't a path to general intelligence--no matter how much text they're fed.

Artificial General Intelligence isn't Here Yet

The impressive success of current "Foundation" or "Frontier" models, such as GPT-4 and Bard, has led some to claim that we are on our way to the promised land: artificial general intelligence (AGI)....

www.jacob-browning.com

Boaz Barak

@boazbaraktcs

9 months

Folks might be a bit too manic-depressive about LLMs with any advance meaning that robot apocalypse is around the corner, and any obstacle means that we've hit the wall. Concretely, pretraining data for LLMs is already basically all of human knowledge, so not being to

29

15

168

9

11

80

Browning.jake00

@Jake_Browning00

1 year

A piece by @ylecun and I. We argue conversation is more than just words; it depends on social norms governing what to say. Current chatbots are oblivious to these norms--and it shows in how dishonest, inconsistent, and offensive they are.

AI Chatbots Don’t Care About Your Social Norms | NOEMA

They seem so human, but they are very alien.

www.noemamag.com

7

24

72

Browning.jake00

@Jake_Browning00

9 months

Why did we think the Winograd Schema Challenge would be a definitive test of common sense? And can there be a definitive test of common sense in language? A piece @ylecun and I wrote back in early 2022 is finally out!

8

68

Browning.jake00

@Jake_Browning00

9 months

I hope that the good social technologies we've developed--and that Andreessen largely lumps into the "enemies"--will help limit some of the dangers of Techno-Optimism. (I don't know what happened to the first version of this post.)

A Blinkered Take on Technology and Capitalism

Manifestos are silly things by nature. A glance at the manifestos of the 20th century reminds us that you need to have an immense ego to speak in the first-person plural and an absurd faith in your...

www.jacob-browning.com

5

11

67

Browning.jake00

@Jake_Browning00

9 months

In a new piece, @AllYouCanPartyand I argue against @add_hawk 's claim that Twitter gamifies communication and transforms our values. We argue Twitter isn't game-like on his account and chasing Likes and Retweets don't seem to transform values. 1/

Why Twitter does not gamify communication

A central question for understanding social media platforms is whether the design of these systems is itself responsible for the harmful effects they have on society. Do these systems push users to...

www.tandfonline.com

3

11

54

Browning.jake00

@Jake_Browning00

2 years

A short piece by Yann LeCun and I. We argue that contemporary large language models, while impressive, are fundamentally limited to a shallow, superficial understanding of the world. This isn't because of the technology, but stem from the limits of language itself.

Noema Magazine

@NoemaMag

2 years

“A system trained on language alone will never approximate human intelligence, even if trained from now until the heat death of the universe.” — @ylecun & @Jake_Browning00

10

65

352

3

9

33

Browning.jake00

@Jake_Browning00

1 year

Really great workshop next weekend, featuring a ton of great speakers and panelists: @ylecun @davidchalmers42 @neurograce @LakeBrenden @luosha @Brown_NLP @raphaelmilliere @glupyan @cameronjbuckner , Nick Shea, and more. Schedule and sign up here:

2

1

19

Browning.jake00

@Jake_Browning00

9 months

@LeszBuk I like Voyager, but memorizing and following scripts of other people accomplishing things in game isn't really planning. If the model did the reverse--learned to play Minecraft, label actions they find useful, and then write a program to accomplish that action--I'd be interested.

1

0

16

Browning.jake00

@Jake_Browning00

6 months

Making language models multimodal doesn't seem to have made them any smarter. And Altman is tamping down hopes that further scaling (or Q*) will lead to transformative AGI. Some reflections on why progress has slowed:

Why isn't Multimodality Making Language Models Smarter?

Philosophy has something called the "symbol grounding" problem. The basic question is whether the words and sentences of language need to be "grounded" in the world through some sort of sensory...

www.jacob-browning.com

3

4

17

Browning.jake00

@Jake_Browning00

10 months

@ylecun As you like to say, people are often caught up in "mathematical hypnotism," getting too wrapped up in the cool math to forget what it is supposed to do.

0

16

Browning.jake00

@Jake_Browning00

10 months

Like phrenology or mesmerism, IIT gained public acceptance with readable and flashy pop-sci presentations by big names rather than empirical investigation. This letter may be a political attempt to change the public conversations, but IIT can't complain; it started it.

hakwan lau 🇺🇦 @hakwan.bsky.social

@hakwanlau

10 months

in light of the spread of some misinformation on public media (including news articles in Nature & Science), a group of 124 researchers weigh in & consider it necessary to label the integrated information theory (IIT) of consciousness as pseudoscience

28

116

396

1

3

15

Browning.jake00

@Jake_Browning00

1 year

Very exciting and provocative new work by @raphaelmilliere on a topic recently taken up by @ylecun @davidchalmers42 @Brown_NLP @glupyan @LakeBrenden and I in our debate at NYU on LLMs and symbol grounding

Raphaël Millière

@raphaelmilliere

1 year

📝New preprint! What does it take for AI models to have grounded representations of lexical items? There is a lot of disagreement – some verbal, some substantive – about what grounding involves. Dimitri Mollo and I frame this old question in a new light 1/

15

84

341

1

5

13

Browning.jake00

@Jake_Browning00

2 years

A reply to our piece from @GaryMarcus But I don't think we agree where he says we do, or disagree on where our disagreements are. But that's the nature of difficult issues, I suppose.

Noema Magazine

@NoemaMag

2 years

Is a decades-long AI debate finally coming to a resolution? @garymarcus is seeing signs that it is. Now "we can finally focus on the real issue: how to get data-driven learning & abstract, symbolic representations to work together."

1

34

197

0

2

9

Browning.jake00

@Jake_Browning00

7 months

Had a really great time chatting with @jjmccammon on his podcast about AI, the limits of LLMs, and why they aren't on their way to personhood.

Will AI ever become a "person?"

Listen to this episode from AI 'N Stuff on Spotify. Full show notes for this episode can be found here: https://www.96layers.ai/p/will-ai-ever-become-a-person Have you ever considered what it truly...

open.spotify.com

2

1

8

Browning.jake00

@Jake_Browning00

1 year

@davidchalmers42 Throw a pitch to @NoemaMag . They're publishing a lot of philosophical takes on AI.

2

0

8

Browning.jake00

@Jake_Browning00

1 year

Audiobook or podcast recommendations for philosophy of AI-adjacent stuff? I have a really long, multiday trip ahead of me.

3

0

7

Browning.jake00

@Jake_Browning00

2 years

A Spanish translation of the recent piece with @ylecun in @NoemaMag ! Thanks to @gienini

gienini

@gienini

2 years

Nuevo artículo de la factoría La IA y los límites del lenguaje Un sistema de IA entrenado solo en palabras y oraciones nunca se aproximará a la comprensión humana Léelo en:

1

2

6

Browning.jake00

@Jake_Browning00

9 months

There's plenty I agree with in @blaiseaguera and Norvig's @NoemaMag piece. But I don't think AGI is here, and I don't think "information tasks" are the right metric for evaluating general intelligence.

Artificial General Intelligence isn't Here Yet

The impressive success of current "Foundation" or "Frontier" models, such as GPT-4 and Bard, has led some to claim that we are on our way to the promised land: artificial general intelligence (AGI)....

www.jacob-browning.com

1

0

6

Browning.jake00

@Jake_Browning00

23 days

@Sander_vdLinden @robsica I don't think that's right. There is a difference between speculation and misinfo. The Convo article is speculation based on public speeches; the only qualified expert on Biden's mental state is his doc. But they've been silent. So everyone is speculating from appearances.

1

0

5

Browning.jake00

@Jake_Browning00

9 months

This doesn't mean there isn't plenty to complain about with Twitter or social media more generally. But we need to distinguish weak claims about ways users can use a technology from stronger claims about how a technology transforms us. 3/

0

5

Browning.jake00

@Jake_Browning00

1 year

@MelMitchell1 @tyrell_turing @VenkRamaswamy I agree, but we shouldn't overstate the power of language. The Eliza Effect occurs even when we know how the program works. Calling it text synthesis might be more accurate but won't help shift perceptions much.

2

1

5

Browning.jake00

@Jake_Browning00

9 months

@danwilliamsphil @add_hawk This is a fantastic piece! We discussed some of these issues while writing, so it is really great to see them really spelled out so clearly.

0

5

Browning.jake00

@Jake_Browning00

7 months

@glupyan Very cool read! The 19th century had its own blue dress moment when color-blindness became public knowledge after a train accident. People also doubted the first reported cases in the 18th c, assuming color-blind folks just never learned their color words.

1

0

4

Browning.jake00

@Jake_Browning00

2 years

@johnschwenkler @NoemaMag is another greater venue for public philosophy

1

4

Browning.jake00

@Jake_Browning00

10 months

If IIT wants to be a science, spend a couple decades in the empirical weeds before publishing the next big, bestselling book.

2

0

4

Browning.jake00

@Jake_Browning00

9 months

@ylecun I complained a bit about the AGI article here, too. If LLMs are AGI, the standards have dropped.

Artificial General Intelligence isn't Here Yet

The impressive success of current "Foundation" or "Frontier" models, such as GPT-4 and Bard, has led some to claim that we are on our way to the promised land: artificial general intelligence (AGI)....

www.jacob-browning.com

1

0

3

Browning.jake00

@Jake_Browning00

2 months

@petemandik @cameronjbuckner @johnmark_taylor Alhazen discussed unconscious inferences in the 11th century. And Pete is right that Leibniz repopularized it in the 18th. But it was super common among the post-Schopenhauer crowd in the 19th century, too. I think Freud was just the best writer of the bunch.

1

4

Browning.jake00

@Jake_Browning00

1 year

@thehangedman In his early Heidegger on Being a Person, he defined authenticity as being consistent among all one's commitments. So you reject or abandon any commitment (e.g., playboy) that is incompatible with the most important (e.g., father and husband).

1

0

4

Browning.jake00

@Jake_Browning00

9 months

We also argue his focus on high frequency posters obscures how the app shapes the experience of the far more numerous consumers. The provocativeness of the analogy with games encourages us to overlook major design choices with troubling effects on users. 2/

1

0

4

Browning.jake00

@Jake_Browning00

1 year

This strikes me as exactly the right point, one that is missing from a lot of the AI risk crowd. They should first prove they can align corporate interests and social good in the present, especially at AI companies, before they talk about uncertain, unpredictable future risks.

Tony Zador

@TonyZador

1 year

"we have an alignment problem, not just between human beings and computer systems but between human society and corporations, human society and governments, human society and institutions." From Ezra Klein's podcast

15

52

228

3

2

4

Browning.jake00

@Jake_Browning00

2 years

@petemandik @gualtieropicc Some concepts survive in a non-scientific capacity, like id and superego, Libra, the passions, hysteria, and so on. But they don't describe anything in current scientific theories.

0

3

Browning.jake00

@Jake_Browning00

9 months

@DavidmComfort @schulzb589 @LeszBuk Snark aside, these aren't action shots; they're atmospheric shots. I want to see what happens next: the zombies attack out of the mist, the calvary crashes into the pikemen, etc. My contention is that AI are failing at complex, difficult-to-describe, multi-object interactions.

1

0

2

Browning.jake00

@Jake_Browning00

23 days

@Sander_vdLinden @robsica I agree about the fake and edited stuff. But isn't Dan's point that the meme isn't misinfo? I feel like the meme wasn't misleading, even at the time. It was a valid worry evidenced by public appearances without any counter evidence. Labeling it misinfo is inaccurate

1

0

3

Browning.jake00

@Jake_Browning00

10 months

@ourwaters @DioVicen Amen. I think scientists often aren't so bad at philosophy as it applies to them. They're just not good with dealing with the weird, a priori arguments and appeals to intuition common in philosophy

1

0

3

Browning.jake00

@Jake_Browning00

23 days

@Sander_vdLinden @robsica But that's on Biden in this case. He could release his cognitive tests like Trump did. People are entitled to infer from Biden's refusal to do so (and, of course, appearances). It seems like misleading propaganda for Biden's supporters to dismiss the issue without evidence.

1

0

3

Browning.jake00

@Jake_Browning00

1 year

It reminded me of a take from 2008 about the iPhone which (rightly) complained that the phone was inferior to a Nokia E70, which had internet, texting, and a much more useful keyboard. I agreed at the time—but, in hindsight, this missed the point. 2/4

1

0

2

Browning.jake00

@Jake_Browning00

1 year

This should be essential reading for anyone trying to make sense of current language models. It does a good job of highlighting why these models are so fascinating, but also fickly, unreliable, and unpredictable.

Sam Bowman

@sleepinyourhat

1 year

I’m sharing a draft of a slightly-opinionated survey paper I’ve been working on for the last couple of months. It's meant for a broad audience—not just LLM researchers. (🧵)

23

275

1K

0

1

3

Browning.jake00

@Jake_Browning00

1 year

@pfau @s_scardapane I think a tiny fraction of the world has gone mad, but they're overrepresented on Twitter.

1

0

3

Browning.jake00

@Jake_Browning00

1 year

I need to write some of this down... Way better than my prepared remarks

iandanforth

@iandanforth

1 year

@raphaelmilliere @Jake_Browning00 @davidchalmers42 @LakeBrenden @ylecun @glupyan @BrownCSDept And here's a preview of the debate courtesy of GPT-4!

1

0

7

0

3

Browning.jake00

@Jake_Browning00

1 year

"Aligning" AI doesn't require more GPUs. It's a social problem requiring independent institutions with teeth. By treating alignment as an in house, technical problem for the future, OpenAI is just making another PR push to avoid governments regulating ChatGPT and Bing AI.

OpenAI

@OpenAI

1 year

We need new technical breakthroughs to steer and control AI systems much smarter than us. Our new Superalignment team aims to solve this problem within 4 years, and we’re dedicating 20% of the compute we've secured to date towards this problem. Join us!

520

744

4K

0

3

Browning.jake00

@Jake_Browning00

2 months

@aresteanu @petemandik You can consent to an animal humping you. That's not a problem. But the law sometimes frowns on this (see Clerks 2).

0

2

Browning.jake00

@Jake_Browning00

10 months

@mjdramstead @ourwaters @DioVicen I mostly saw philosophers telling scientists they didn't know what science is and thus had to treat phrenology and mesmerism as valid sciences, too. There were also moral claims that it was wrong to exclude ideas and theories that had been repeatedly falsified.

2

0

2

Browning.jake00

@Jake_Browning00

1 year

This is fantastic.

Subbarao Kambhampati (కంభంపాటి సుబ్బారావు)

@rao2z

1 year

Our new paper generalizing the chain, circle and graph of thought prompting strategies--that unleashes the hidden power of LLMs (and graduate students). Hope @_akhaliq picks it up.. 🤞

12

40

248

0

2

Browning.jake00

@Jake_Browning00

1 year

@xgabegottliebx I was in a bar the other night called "the Emerson." It had busts of the great American poets and, weirdly, there was Holmes. Dude couldn't pick a lane.

1

0

2

Browning.jake00

@Jake_Browning00

1 year

@DeonTBenton @matthewslocombe This looks great! I'll make that my next book

0

2

Browning.jake00

@Jake_Browning00

10 months

@mjdramstead @ourwaters @DioVicen IIT's physical claims has been disproven, multiple times for each iteration of the theory. But no one can disprove the phenomenological part, which is treated as "axiomatic." Which is the problem. The details of the theory keep changing because the core is impervious to evidence.

1

0

2

Browning.jake00

@Jake_Browning00

1 year

@schafer_karl This is part of Douglas Hofstadter's argument in his book on analogical reasoning. And Melanie Mitchell develops it some in her book on AI.

1

0

2

Browning.jake00

@Jake_Browning00

2 years

@AntonellaTrama3 Second @martacarava . Haugeland's definition is often used (e.g., by Clark and Toribio): 1) reps store info about environment, 2) can be used to guide even when what it's representing isn't present, and 3) is part of a systematic representative schema of the organism.

1

0

2

Browning.jake00

@Jake_Browning00

1 year

@SeeReason @perrymetzger @schulzb589

0

1

Browning.jake00

@Jake_Browning00

1 year

@DeonTBenton Is there any text or texts that does anything as comprehensive as Spelke's last book? Since I don't care for her explanations of findings, I'd love to read a good contrary account.

1

0

2

Browning.jake00

@Jake_Browning00

2 years

@keithfrankish @Philip_Goff Matthias Michel is doing great work on the history of scientific work on consciousness.

1

0

2

Browning.jake00

@Jake_Browning00

1 year

Smart article. Rather than criticize x-risk crowd for their beliefs, instead criticizes their attempts to circumvent existing institutions. Strengthening these institutions' ability to address current problems is essential for dealing with any future risk

Panic about overhyped AI risk could lead to the wrong kind of regulation

There’s something missing in the heart of the conversation about AI.

www.vox.com

1

0

2

Browning.jake00

@Jake_Browning00

9 months

@JudiciaIreview Part of it is they are running out of data and using synthesized data compounds errors. So scaling up won't bring benefits.

2

0

1

Browning.jake00

@Jake_Browning00

2 years

@GaryMarcus @cameronjbuckner @ylecun @davidchalmers42 @NoemaMag I'm just not sure I can imagine domain-general nativism, so it's hard to argue against it

2

0

2

Browning.jake00

@Jake_Browning00

2 years

@petemandik LaMDA includes a specially trained module which scores potential responses to pick out the most appropriate to spit out in a way similar to how CLIP scores potential images made in dall-e to provide the best. It's not introspection but it isn't unrelated

1

0

2

Browning.jake00

@Jake_Browning00

21 days

Ah the question of determinancy. To reappropriate Chisholm's critique of Ayer: if you dream of a tiger, does it have a determinate number of stripes? Try and imagine one with just 17 stripes.

Balázs Kégl

@balazskegl

21 days

When you dream, do you see images or just imagine abstract representations?

12

0

4

0

2

Browning.jake00

@Jake_Browning00

2 years

@AntonellaTrama3 @martacarava Bermudez, in his cog sci textbook, goes with the more generic: "[representations are] stored information about the environment."

0

2

Browning.jake00

@Jake_Browning00

7 months

@glupyan Dalton's study on his own color-blindness (the first widely believed study) of this is a good example. He couldn't figure out why his fellow botanists disliked the way he described flowers, so he finally tested their color vision and realized he was the odd duck.

1

0

2

Browning.jake00

@Jake_Browning00

1 year

Really good read on the dangers of Google's AI Search and its potential effect on news creators. Plagiarism Engine: Google’s Content-Swiping AI Could Break the Internet

Plagiarism Engine: Google’s Content-Swiping AI Could Break the Internet

The Search Generative Experience seems more like a text-copying experience.

www.tomshardware.com

0

1

2

Browning.jake00

@Jake_Browning00

2 years

Brilliant work from my colleague Philipp Schmitt!

Philipp Schmitt

@philippschmitt

2 years

New research-y project: Blueprints for Intelligence, a visual history of artificial neural networks from 1943 to 2020

41

518

2K

0

2

Browning.jake00

@Jake_Browning00

9 months

@DavidmComfort @schulzb589 @LeszBuk I love all the horses moving in the wrong direction. But at least they aren't clipping each other.

1

0

2

Browning.jake00

@Jake_Browning00

1 year

Turns out there is no overlap between my Twitter use and my Instagram use.

1

0

1

Browning.jake00

@Jake_Browning00

2 years

@GaryMarcus @cameronjbuckner @ylecun @davidchalmers42 @NoemaMag Maybe we should have argued that, but we were actually targeting domain-specific symbol manipulation--that some modules in the brain come innately with discrete symbols, variable-binding, etc.

2

0

1

Browning.jake00

@Jake_Browning00

1 year

@KhurrumM @ylecun @LakeBrenden @glupyan @davidchalmers42 Bad example. She had senses and was embodied for years before she learned language. Also had a wicked sense of humor.

1

0

1

Browning.jake00

@Jake_Browning00

1 year

@carl_b_sachs And my "Pittsburgh School" critique: .

0

1

Browning.jake00

@Jake_Browning00

2 years

@WiringTheBrain Can I get a copy of the reading list for this project? It sounds amazing to me.

1

0

1

Browning.jake00

@Jake_Browning00

6 months

@TossingsRaphael I think model free approaches are pretty hopeless. But mixed self-supervised and RL approaches, like Nvidia's work on Minecraft and NetHack, seems very promising.

0

1

Browning.jake00

@Jake_Browning00

2 years

@GaryMarcus @cameronjbuckner @ylecun @davidchalmers42 @NoemaMag Yann's proposal was designed to generate and evaluate abstract representations of plausible future states based off the present. And it doesn't assume any discrete symbols or variable binding.

1

0

1

Browning.jake00

@Jake_Browning00

5 months

"Members of the public should continue to be skeptical, but not cynical, while asking for better evidence and arguments before reflexively accepting a reported scientific consensus."

Who Should You Trust? Why Appeals to Scientific Consensus Are Often Uncompelling

Consumers of scientific information should be skeptical of an apparent scientific consensus. Consider: How politicized is this topic? What are the career incentives for the scientists? How easy would...

www.skeptic.com

0

1

Browning.jake00

@Jake_Browning00

1 year

@MelMitchell1 @tyrell_turing @VenkRamaswamy This also reminds me of the excellent short piece by @kmahowald and @neuranna

0

1

Browning.jake00

@Jake_Browning00

2 years

@WiringTheBrain @cameronjbuckner @GaryMarcus @ylecun

Deon T. Benton

@DeonTBenton

2 years

Ever wonder what would happen if a nativist and an empiricist were to meet? You'll soon get that answer because my friend (and nativist!), Dr. Jenny Wang ( @JinjingJenny1 ), will be joining me to co-host @TheItsInnatePC and to debate the origins of human knowledge and concepts!

5

6

63

0

1

Browning.jake00

@Jake_Browning00

2 years

@rgblong @GaryMarcus @cameronjbuckner @ylecun @davidchalmers42 @NoemaMag And is a really clear paper!

0

1

Browning.jake00

@Jake_Browning00

3 years

@NoamChompers Weirdly, Kant uses it in a letter in 1783 to Marcus Herz.

0

1

Browning.jake00

@Jake_Browning00

1 year

@bleepbeepbzzz Maybe I'm mislabeling, but I'm thinking of (for example) Stuart Russell, the idea that the Midas problem is the big issue, and inverse RL as a potential solution. I'm not thinking of, like, McKinsey risk analysis.

1

0

1

Browning.jake00

@Jake_Browning00

1 year

@raphaelmilliere @ylecun @davidchalmers42 @Brown_NLP @glupyan @LakeBrenden I remember you mentioning this piece afterwards! Glad it is finally out!

0

1

Browning.jake00

@Jake_Browning00

2 years

@MilekPl Pamela McCorduck is a good reference, since she wrote the early history of AI:

1

0

1

Browning.jake00

@Jake_Browning00

9 months

@JudiciaIreview In principle, maybe. In practice, Generative AI hallucinates so much and so unpredictably that you'd end up steadily deviating from reality. And we actually already see that happening

0

1

Browning.jake00

@Jake_Browning00

10 months

@marshall5912 @nikstankovic_ @whyvert @BrankoMilan @DanielCanueto Don't forget the forced sterilizations!

0

1

Browning.jake00

@Jake_Browning00

1 year

The iPhone may not have been a great phone, but it was an incredible PC, with an enclosed ecosystem that transformed the design, purchase, acquisition, and usage of programs (relabeled “apps”). Apple defined the smartphone and owns the market. 3/4

1

0

1

Browning.jake00

@Jake_Browning00

9 months

@Shawnryan96 @ylecun That is a great article and a huge accomplishment. I think that will help with a lot of problems. But it can't fix the limits of the autoregressive approach to language.

1

0

1

Browning.jake00

@Jake_Browning00

1 year

This is true regardless of what future risk means. I'm skeptical it is existential but, if it is, building up current institutions and ensuring current harms from AI are addressed is still necessary for ensuring capacity to deal with whatever comes next.

0

1

Browning.jake00

@Jake_Browning00

1 year

(Sent from an iPhone)

0

1

Browning.jake00

@Jake_Browning00

2 months

@marielgoddu Burnyeat is really critical of the functionalist reading of Aristotle for the reasons you mention. He argues our conception of function is only possible when Descartes regards life as mechanical. Jessica Riskin's book is about the merging of mechanical and biological teleology.

0

1

Browning.jake00

@Jake_Browning00

2 years

@petemandik Blaise Aguera y Arcas makes that case.

Do large language models understand us?

LLMs have a great deal to teach us about the nature of language, understanding, intelligence, sociality, and personhood.

link.medium.com

0

1

Browning.jake00

@Jake_Browning00

2 years

@cameronjbuckner @NoemaMag We also linked to your piece for Junkyard of the Mind (because that piece is fantastic).

1

0

1

Browning.jake00

@Jake_Browning00

1 year

@HusarenH @NoemaMag @ylecun Yes, but you can get a chatbot to apologize and admit it was wrong.

0

1

Browning.jake00

@Jake_Browning00

9 months

@MaunaLoona @ylecun Sure, but feeding more modalities to a language model still treats the core of thought as language. Better success, I think, will come from building a better, multimodal world model and then adding a language modality to it--like evolution did with hominids.

1

0

1

Browning.jake00

@Jake_Browning00

11 months

@AISafetyMemes @cellinip @ylecun @NoemaMag I'd reconsider if an LLM could drive a car cross country. Humans are pretty good at that and it turns out to be really cognitively complex but not dependent on language.

1

0

1

Browning.jake00

@Jake_Browning00

10 months

@GaneshNatesh @mjdramstead @ourwaters @DioVicen @JohannesKleiner Don't have to convince me! I've always said the same. That's why treating them as axiomatic is so frustrating.

1

0

1

Browning.jake00

@Jake_Browning00

1 year

In fairness, it is easy to underestimate the novel: probably the first hammer seemed fragile and harder to use than a nice rock. But you should be careful about judging new tech by old metrics. 4/4

1

0

1

Browning.jake00

@Jake_Browning00

1 year

How might telepathy actually work outside the realm of sci-fi? – via @aeonmag

0

1

Browning.jake00

@Jake_Browning00

1 year