Jonas Metzger @JonasMtzgr
I try to compute economic problems away. PhD Economics and PMn Computer Science candidate at @Stanford. Palo Alto, CA Joined February 2018-
Tweets290
-
Followers325
-
Following849
-
Likes1K
"In a first-ever human clinical trial of four adult patients, an mRNA cancer vaccine developed at the University of Florida quickly reprogrammed the immune system to attack glioblastoma, the most aggressive and lethal brain tumor." medicalxpress.com/news/2024-05-m…
Philosophically, what's the dividing line between Icarus and the Wright Brothers?
agents started as fun sideprojects a few months ago, now they're key to state of the art performance
agents started as fun sideprojects a few months ago, now they're key to state of the art performance
Will AI cause mass unemployment? It's not a scientific question, it's a design problem. If we build systems to be operated by + amplify *current* workers there won't be mass unemployment, if we build systems that displace them there will be. Both are difficult and distinct design…
Will AI cause mass unemployment? It's not a scientific question, it's a design problem. If we build systems to be operated by + amplify *current* workers there won't be mass unemployment, if we build systems that displace them there will be. Both are difficult and distinct design…
Sharing a bit more about Reflect Orbital today. @4TristanS and I are developing a constellation of revolutionary satellites to sell sunlight to thousands of solar farms after dark. We think sunlight is the new oil and space is ready to support energy infrastructure. This…
System 03 is deployed in the Great Pacific Garbage Patch again, for the first time in 2024, as we go into our most ambitious year yet. (1/4)
Anecdotal evidence for an issue AI safety researchers raised for a while. Self-supervised learning yields models that pick up on subtle clues that they are being evaluated.Implies that current RLHF methodology is insufficient for aligning the systems we'll build in the next years
Anecdotal evidence for an issue AI safety researchers raised for a while. Self-supervised learning yields models that pick up on subtle clues that they are being evaluated.Implies that current RLHF methodology is insufficient for aligning the systems we'll build in the next years https://t.co/LjvjFyZMXP
This is one of the few problems with AI that *get worse* at scale (=one we should actually worry about). Increasing model capacity makes it feasible/optimal to "pretend" to be aligned within the RLHF domain, while remaining pretraining-loss-optimal outside the RLHF domain.
This is one of the few problems with AI that *get worse* at scale (=one we should actually worry about). Increasing model capacity makes it feasible/optimal to "pretend" to be aligned within the RLHF domain, while remaining pretraining-loss-optimal outside the RLHF domain.
Chinese President Xi has called for the mass production of humanoids by 2025. cbsn.ws/4bKlE4B
We now know how to teach AI a causal world model by letting it watch online videos. That's right: it infers a latent action space without seeing any ground-truth actions. It's hard to overstate what how this will play out over the next years. I'm calling generally intelligent…
We now know how to teach AI a causal world model by letting it watch online videos. That's right: it infers a latent action space without seeing any ground-truth actions. It's hard to overstate what how this will play out over the next years. I'm calling generally intelligent…
Unbelievable: the North Atlantic sea surface temperature is now 4.5 standard deviations above the recent 1991-2020 climatological mean... that translates to a 1-in-284,000-year event. Yet here we are watching it unfold, one day at a time. This is deeply troubling.
anyone remember the 2018 "Large-Scale Study of Curiosity-Driven Learning" paper calling it 'a limitation' that their agent gets stuck in front of a TV? This was literally baby AGI telling us it needs to do self-supervised learning on videos
Nothing to see here, definitely no reason to think trying to make LLMs “safe” by training them extensively on behaviors-to-avoid will backfire
Nothing to see here, definitely no reason to think trying to make LLMs “safe” by training them extensively on behaviors-to-avoid will backfire
significant milestone
to me, findings like these suggest we may think of LLMs (&multimodal successors) as a discovery, rather than as engineered objects. we try to approximate/study/tame a platonic ideal: "the neural net that is least surprised by the next bit on the internet". the design space is…
to me, findings like these suggest we may think of LLMs (&multimodal successors) as a discovery, rather than as engineered objects. we try to approximate/study/tame a platonic ideal: "the neural net that is least surprised by the next bit on the internet". the design space is…
A compelling intuition is that deep learning does approximate Solomonoff induction, finding a mixture of the programs that explain the data, weighted by complexity. Finding a more precise version of this claim that's actually true would help us understand why deep learning works…
If SSMs like Mamba really replace transformers, then prompting LLMs will become funny. Before reading documents, the model has to be told what it's reading them for, else it will pay attention inefficiently. There will be lots of re-reading prompting techniques. And value in…
If SSMs like Mamba really replace transformers, then prompting LLMs will become funny. Before reading documents, the model has to be told what it's reading them for, else it will pay attention inefficiently. There will be lots of re-reading prompting techniques. And value in…
MaureenJoan @r3x6NQmQ0I4pkOh
2 Followers 373 FollowingJeanLucas @4f10ePsQnK2O4TY
0 Followers 224 FollowingLaurenz De Rosa @LaurenzDeRosa
2 Followers 97 FollowingAnnalise Sellards @SellaAnnal
85 Followers 5K FollowingKairtir @kairtir68223
1 Followers 326 FollowingVickyJerry @O97CbiQSkPvo0c
12 Followers 383 FollowingTheanew @TheanewTLY5
0 Followers 191 FollowingMarion Menton @MarionMent41046
47 Followers 5K FollowingPresley Castellonese @PCastellon62847
80 Followers 5K FollowingDrausau @Drausau_Ah
0 Followers 266 FollowingPhil Morettini @TechnologyGuy
16K Followers 11K Following IT Industry C-Level Interim Exec, Consultant & Blogger to SaaS, Software & Hardware Companies. Loves the Illini, Cubs & Bears: https://t.co/uDUzDbg5YnWendyOwen @g79h577K73T3Y
0 Followers 213 FollowingEleanoreLandon @1WKmjWavDF92D
1 Followers 367 FollowingMcToshe @TosheMc35940
0 Followers 325 Followingきだゆか @kidayuka151960
1 Followers 399 FollowingTheser @Theser271420
39 Followers 707 FollowingRosemaryLew @s0K00gVaK6mUr
2 Followers 369 FollowingJaniceAlfred @M0Vw9fG8dYPhh
0 Followers 385 FollowingIndoor Houseplant @annas_plant
79 Followers 189 Following Just another over watered houseplant. Photosynthesising one tweet at a timeMartin Åström @Martin_Astrom
38 Followers 1K Following Mainly tryin’ to be funny. Doin’ so-so. Guess you’re laughing on the inside?Colleen 😍 @Colleen__L9
3 Followers 480 Following Рrovocativе sirеn driven bу an unquеnchаblе thirst for рleasureTrevor Loy @trevorloy
18K Followers 3K Following VC investor emerging ecosystems @FlywheelVC. Lecturer entrepreneurship & VC @Stanford. Prev: BoD @NVCA; Mentor @KauffmanFellows; 3x founder; Chip design @Intel.Minh Nhat Nguyen @menhguin
501 Followers 2K Following AI Safety research+climate protests. Owner of @AIHubCentral. Product @hume_ai. Don't do the deferred life plan.Sophia @isarida83831211
2 Followers 932 FollowingJack Kanteh @kanteh_jack
269 Followers 2K FollowingHildred Wygant @HWygant90581
64 Followers 5K FollowingSierra Usry @SierraUsry38994
87 Followers 5K FollowingJoe Paul @arg_maximus
20 Followers 165 FollowingKerry Paterniti @PaternitiK88800
90 Followers 5K FollowingDigna Tidwell @DignaTidwe89375
89 Followers 5K Following 20 , New Jersey , Digna🍚 , TOP0.69% onlyfansNatashia Carignan @CarignNatas
25 Followers 5K FollowingSophia Kazinnik @SophiaKazinnik
351 Followers 1K Following Richmond Fed | Views my own https://t.co/wzywF1lMW6Marcelo Ortiz-Villavi.. @Marcelortizv
452 Followers 795 Following 🇪🇨 Econ. Ph.D. Student @EmoryEconomics | Econometrics, Causal Inference and Machine LearningMaryjane Ardaly @arda_maryj
89 Followers 5K FollowingDavid Atkinson @diatkinson
128 Followers 1K Following PhD student @KhouryCollege, interested in AI interpretability. Previously @EpochAIResearch.China Mavle @mav_chin
28 Followers 5K FollowingGautam Machiraju 🌺 @gmachiraju
659 Followers 5K Following PhD-ing @StanfordAILab w/ @ParagMallick @HazyResearch🌲 AI-driven data copilots for scientific discovery♟️🧬🔬🛰🔭 Powered by prog house, people, 3rd places 🪩✨RicardaMorillo @morillo92509
113 Followers 2K FollowingOliver Hsu @oyhsu
5K Followers 971 Following Partner @a16z investing in American Dynamism 🇺🇸. Frontier technology & culture. Previously @mosdotcom @andela.EB @Emarl_bee
922 Followers 661 Following Creating @Stanford and @google X | Dance | Fiction | Stanford DS Scholar | Chevron Energy Fellow | ML and Optimization for Energy x Climate. #LFCIsaiah Taylor — (�.. @isaiah_p_taylor
10K Followers 892 Following Founder @valaratomics. Kardashev scale climber, Malthus disrespector, Age of Man enjoyer. Ask me why consuming energy is good and you should have more babies.benahorowitz.eth @bhorowitz
652K Followers 607 FollowingBen Nowack @bennbuilds
6K Followers 183 Following We sell sunlight after dark. https://t.co/BAN6JRW1KKEric Steinberger @EricSteinb
7K Followers 479 Following Writing code that writes code on a mission to build safe superintelligence | CEO/cofounder @magicailabsZvi Mowshowitz @TheZvi
24K Followers 283 Following Blogger world modeling, now mostly AI and AI x-risk, at Don't Worry About the Vase (https://t.co/O9LbMQjKoo or WP/LW), founding Balsa Research to fix policy.Rain @Rain__AI
2K Followers 30 Following We are building artificial brains to power the most important AI applications in the future.Daniel Mason @dgmason
5K Followers 2K Following Founder @anonplatform 🔌🌐 Before: partner @hiframework & founder @springlabsTommy @thomasehendrix
1K Followers 515 Following General Partner @decisivepointvc | Prev: @Blackstone & Green Beret @USSOCOM | #investedinthemission | Views are my own.Amnon Shashua @AmnonShashua
7K Followers 209 Following CEO @Mobileye. @OrCam, @AI21Labs, @ONEZEROBANK. Sachs Prof. Computer Science at Hebrew U. Passion for #AI, cycling and #SelfDrivingCars. Opinions are my ownGabriel Mukobi @gabemukobi
343 Followers 318 Following Fellow @RANDCorporation, CS PhD @Berkeley_AI | AI Governance, Safety, and Alignment. Opinions are my own.RoboDepot🤖 @RoboDepot
711 Followers 2K Following Your ultimate nexus for robotics, AI, ML, CV, research papers, small/DIY robots and future tech showcases. Follow to see the robotics revolution!🔔Matt Freed @_mattfreed
2K Followers 598 Following To live in the future, we first have to build it | making lil’ things like embodied AGI @kscalelabs | MechE & AI 🇺🇸Noémie Federico @FedericoNoemie
802 Followers 367 Following Helping Founders build in the US @beondeck | O1 extraordinary alien 🛸🇺🇸🦅| culturally confused, linguistically challengedZanë ([email protected].. @ZanaBucinca
2K Followers 404 Following PhD Candidate @Harvard, human-AI interaction; KosovoMinh Nhat Nguyen @menhguin
501 Followers 2K Following AI Safety research+climate protests. Owner of @AIHubCentral. Product @hume_ai. Don't do the deferred life plan.Visual Studio Code @code
686K Followers 133 Following Microsoft Visual Studio Code lets you build and debug modern web and cloud applications. Visual Studio Code is free and available on Linux, macOS, and Windows.Sophia Yang, Ph.D. @ .. @sophiamyang
18K Followers 1K Following Head of Developer Relations @MistralAIAlexander Koch @alexkoch_ai
5K Followers 209 Following Founder of Tau Robotics (@taurobots) | Z Fellow | Emergent Ventures Fellow 2024Taelin @VictorTaelin
18K Followers 921 Following Founder of @HigherOrderComp Building the massively parallel future of computing Reaching AGI to cure all diseases and suffering is all that mattersstephen balaban @stephenbalaban
9K Followers 1K Following Co-founder, CEO - Lambda. We’re hiring: https://t.co/n0yFaq0ZDnChelsea Finn @chelseabfinn
69K Followers 384 Following Asst Prof of CS & EE @Stanford. PhD from @Berkeley_EECS, EECS BS from @MITLighthouse @lighthousehq_
567 Followers 4 Following Betting on America 🇺🇸 We help the world's best and brightest get a US work visa, fast, so you can get back to buildingMatt Clifford @matthewclifford
25K Followers 2K Following Co-founder @join_ef; Chair @ARIA_Research; co-led AI Safety Summit at Bletchley ParkNoland Arbaugh @ModdedQuad
87K Followers 22 Following P1. Cyborg. Neuralnaut. Skynet Progenitor. Taurus.stacy 🌤 @voidshapes
1K Followers 492 Following lifelike automata 🌗 mutation hunter working on aging & longevity / phd candidate @ucberkeley. I do all my own stuntsTamay Besiroglu @tamaybes
3K Followers 722 Following Thinking about economics, computing and machine learning @EpochAIResearch @MIT_CSAILringwiss @ringwiss
18K Followers 174 Following 🏳️🌈 🇪🇺 🇵🇱 🇬🇧 He/him. Armchair parliamentarian. I type at 140 wpm. RTs are endorsements.Hyperstition, Inc @hyperstition_x
4K Followers 1 Following The future is forged by the hammer of cognition against the anvil of creation.Patrick Hsu @pdhsu
30K Followers 3K Following co-founder @ArcInstitute, @BerkeleyBioE professor | researcher & investor in genome editing, biotech, AI | 🇨🇦 @harvard @broadinstitute @editasmed, Fast GrantsPhysical Intelligence @physical_int
4K Followers 8 Following Physical Intelligence (Pi), bringing AI into the physical world.Cognition @cognition_labs
125K Followers 19 Following Makers of Devin, the first AI software engineer. We are an applied AI lab focused on reasoning, and code is just the beginning. Join us: https://t.co/tpfZwEwGiqEdouard Harris @harris_edouard
5K Followers 2K Following Cofounder & CTO @GladstoneAI https://t.co/aGM324ATKGAugustus Doricko @ADoricko
11K Followers 982 Following Making 💦 WATER ABUNDANT💦 @ RAINMAKER. DM to discuss deadlifts #beastmodeOliver Hsu @oyhsu
5K Followers 971 Following Partner @a16z investing in American Dynamism 🇺🇸. Frontier technology & culture. Previously @mosdotcom @andela.Varda Space Industrie.. @VardaSpace
28K Followers 41 Following Building the world's first industrial space manufacturing platform and earth re-entry logisticsGeneral Fabrication, .. @genfabco
2K Followers 0 Following We make machine tools to enable permanent industrial revolution.Søren Kierkegaard | .. @Kierkegaarddd
42K Followers 23 Following Quotes by Søren Kierkegaard | Danish Philosopher, Poet & Author ✍️ | Think Better, CLICK 👉 https://t.co/WdL040zeG9Houda Nait El Barj @Houda_nait
398 Followers 692 Following PhD Econ @Stanford I believe technology can create a better world v=sqrt(2gh) our existential equation for noumenal error Econ/Probability/AI/PhilosophyGabriel Filsinger @GabrielFilsing1
181 Followers 340 Following Post-Doc Michael Fischbach's lab @mfgrp @Stanford Ph.D. @geochurch lab, @Harvard Systems BiologyAaron Erickson @AaronErickson
4K Followers 3K Following Sr. Engineering Manager at @NVIDIA building autonomous observability agents. Views are mine, not of employer. ex-New Relic/Salesforce/ThoughtWorksSampriti Bhattacharyy.. @sampritibh
7K Followers 341 Following CEO & Founder @navierboat Roboticist🤖. Ex @ NASA🚀. MIT PhD'17. Building electric flying boats 🌊🚤I find it funny when people complain about “difficult people”. I feel like my whole life has been focused on how to work best with “difficult people”. It’s my professional moat. In my experience, all truly capable people are difficult in some way. Deal with it!
This paper shows when, why, how, and how much trade agreements can help in motivating environmental conservation, instead of exploitation, if a typical agreement is modified so that tariffs can be contingent on resource stocks (/forest cover). (It builds "Trade and Trees", AER:I)
Although free trade can lead to deforestation, tariffs that are contingent on the forest cover, where default tariffs can vary with changes in the resource stock, can motivate conservation, from @BardHarstad nber.org/papers/w32392
The US network admits me because I have arrived while the Europeans because I left -- it appears🙄 (Having returned to the US, after a decade in Europe, I’m readmitted to NBER and, on the next day, CEPR.) cepr.org/about/people/b… nber.org/people/bard_ha…
Don't be a B-lane swimmer.
Years ago @vboykis sent me a wonderful essay on exactly this. Sadly, it's only available on the internet archive now: web.archive.org/web/2024031317…
@MrEdogtagnft On Monday, I worked from 9am Beijing time to 2am Texas time, but to each their own
btw, this is the thing that makes current meat
Thrilled that a paper from my PhD, "Faster Maximum Inner Product Search in High Dimensions" has been accepted to ICML 2024! In the paper, we accelerate state-of-the-art for the Maximum Inner Product Search (MIPS) problem. MIPS is a core subroutine in systems like recommendation…
What happens when generative models are trained on their own outputs? Prior works foretold of a catastrophic feedback loop, a curse of recursion, descending into madness as models consume their own outputs. Are we poisoning the very data necessary to train future models? 1/N
New research post on refusals in LLMs lesswrong.com/posts/jGuXSZgv…
@AndreTI @SpencrGreenberg @max_spero_ I think on timescales approaching decades, limiting yourself to predictable advances has a very poor track record, since it systematically hugely underestimates innovation. Major, 'unpredictable' advances happen from time to time, and that fact is predictable
A professor retweets a student who has made a political statement pro his side and says “these are the kinds of students Columbia should be striving to admit.” Claims like this should immediately disqualify a faculty from any admission decision.
After two years in stealth mode, we're thrilled to unveil Mentee Robotics and our humanoid robot, @MenteeBot! With AI integration at every layer, from Sim2Real machine learning to NeRF-based algorithms and LLMs, we've achieved a complete end-to-end cycle for tasks. This is just…
@datepsych You can tell a person is smarter than you by observing that they continuously make poor choices that nevertheless work out in their favor by what appears to be luck.
Wishing all people prosperity, joy, and peace this Passover. For many, it is a seder interrupted, and we wish for an end to war, to hate, to death, and we pray for the safe return of hostages. ☮️❤️
.@ClintonGlobal has done incredible work for global water infrastructure It was a pleasure discussing how cloud seeding can enhance water supplies with # 42 @BillClinton !
Open source CAD model here: cad.onshape.com/documents/71f7…
Won’t be the last time we get copied If it’s not obvious yet, Figure is doing the best mechanical design in the world for robotics Let us cook