@rtk254
Researcher & entrepreneur | Building a sensemaking layer for research @CosmikNetwork | Stigmergic cognition | Prev @AsteraInstitute
๐ Joined: 11/21/2018
๐๏ธ Archived: 10/17/2024
3.8K Tweets
2.0K Followers
3.6K Following
24.6K Likes
Downloads the complete Twitter archive in JSON format
| Username | Mention Count |
|---|
Loading tweets...
Selectively breeding the most *individually* productive hens leads to lower overall egg production, so chicken farmers have cleverly learned to shift selection to the level of the *group*. Economists, on the other hand > https://t.co/TfNtXqGL9x
have achieved a degenerate reversal of devaluing everything except personal gain. When will the chicken farmers rescue us from the neoliberal economists? https://t.co/mZwjjWct7W
@ubiquity75 Intersting recent research on this! https://t.co/aTVcIK1jEY
@hermittoday โWe would rather be ruined than changed. We would rather die in our dread than climb the cross of the moment and see our illusions die.โ -- W.H. Auden
๐ filing under "we already have superintelligences with anti-human goals - corporations" https://t.co/yGMQnpIWT1
sources : https://t.co/lidmqyjRQm https://t.co/3nxHPSGFZk
GPT4 - veils are off, current big tech NLP losing any pretense of academic research: https://t.co/aEEJnXSwvc
Cool timeline by @jhamrick on AI and cog-sci converging/diverging over the years, hopefully starting to come back together! :) #BAICS2020 #ICLR2020 https://t.co/rDnReqfsOq
๐Life update: I'm starting a new position as an Entrepreneur-in-Residence @AsteraInstitute! Iโll be working on upgrading science social media by combining AI and federated semantic nanopublishing to improve how we collaboratively make sense of science (in the open!) > https://t.co/AJzhPFMwCR
https://t.co/Osrgms2gKv https://t.co/EsrAc3Sy2L
A few months late - my PhD Thesis is finally up on arXiv ๐ "What's my model inside of?": Exploring the role of environments for grounded natural language understanding TL;DR in two excerpts https://t.co/5dYfVW712v #PhDone https://t.co/cYtQwryCRW
1/Excited to share some new work on collective sensemaking! Weโll present it in the Blue Sky Ideas track (~open, exciting areas for R&D) of the upcoming @ACMHT This is a collective effort of @daostack, Veeo, @InferenceActive @CSenseMakers @HyadataLab https://t.co/JAcnB4YSpk๐งต
Are there any good theories about why this happens? E.g., why does the tech focus on automating creative work as opposed to automating the really tedious things? Is it just since the creative work makes for flashier headlines? I feel like there is something deeper here... https://t.co/vIUZDGbTLQ
New year, new ways to do science! ๐พ๐ญ I'm excited to share some details about Sensemaking Networks, the project Iโll be developing this year @AsteraInstitute! ๐งต1/ Blog post version (2 parts): https://t.co/luSwcHXS1k https://t.co/eZ0ysMt7mJ https://t.co/ceLVkDRnbQ
@lizweil Thanks, looking forward to reading this! One thought meanwhile - the fact that we have to delve deep into the psyches of these self-elected "leaders" is a sign of a deeply FUBAR situation. @doctorow writes about Zuckerberg in "The Internet Con" and the same holds for Altman https://t.co/o8WF4awRV2
Lakoff's audacious Conceptual Metaphor Theory becoming more grounded: "Convergent evidence from neural, cognitive, and information sciences is pointing toward a fascinating hypothesis: (1/n) https://t.co/Jf0WleIE74
@mmbronstein The processes that shape what research questions are asked, how they are pursued, what research gets attention, all of these are political. See research by @yoginho @evantthompson for example below https://t.co/HnbYCVeCaT https://t.co/RY8eY3JXeh
Excited to share our new preprint! We take a deep dive into the bAbI benchmark. Why bAbI? It targets a core NLU competency of *grounded situation modelling* for which very few benchmarks currently exist https://t.co/QDygVlKtG4. Longish ๐งต> 1/ https://t.co/WIsC8e2KBz
๐ https://t.co/qa1C7tvqvF "we contacted the service while undercover as a fictional author, and managed to purchase 50 citations" hardcore investigative academia ๐ต๏ธ h/t @shirandudy
๐ข Hey #ScienceTwitter! Would you join an open science social media mirroring service? It would ensure that all your research-related posts were citeable & open access, regardless of source platform (X, Masto, Bsky,)! ๐ RT if youโre interested ๐ย Also - vote below for featuresโฆ
Super excited to start a visiting research program this year at @judyefan's Cognitive Tools (https://t.co/En8tWyYs0l) lab in UCSD! Looking forward to working on questions at the intersection of cognitive science, comp. ling. and AI. And also sea and sun ๐๐
Interesting paper by Marco Baroni on evolving useful languages. https://t.co/KUpdUm9BdH 1. "Formal semanticists with a heart condition might want to stop reading at this point." lol 2. Focus should be on designing environments in which agents develop properties of interest
Open access to information != open access to tools and capacities to *make sense* of information Some related research of mine with @InferenceActive : https://t.co/fDSLPZOwSo https://t.co/bJ7C93CxOM https://t.co/rGJNgTk0UX
The semantic web phoenix is rising from the ashes https://t.co/npXfJVJHD1 https://t.co/ndRYnA2igO https://t.co/82lshZfsuS
For all the impressive performance of LLMs like ๐/๐ฆฉ, I would still be really surprised if they could reliably solve the type of challenge we suggested in our ACL2020 paper, eg a story + qs about the final state (how many fruits are now in crate 2?) https://t.co/rpIoa76MTX https://t.co/iE9LXYXoYk https://t.co/pzoz3jDPTS
Interesting paper - "by introducing position tokens (e.g., "3 10e1 2"), transformers learn to accurately add and subtract numbers up to 60 digits" Another angle - exposing model to "meaning behind the symbols" does wonders for performance. How to do this in NLU as well? https://t.co/eS7l9cRbzB
@JimRuttShow this might be related to our conversation on why our current economic system selects for sociopaths https://t.co/ArETV06pdi
Iโll be presenting Sensemaking Networks next week at @ICSSIConference! ๐Social media is part of the "dark matter" of science research, exerting dramatic impact but off the scientific record ๐ฆ We'll talk about how to make it visible ๐ Also - Iโd love to meet up with fellow sensemakers in DC, feel free drop me a line if youโre around :) (extended abstract in first reply -> )
Linguistic Embodiment Hypothesis: language as an embodied experiential modality like perception, etc, but also shared and public (eg Clark's extended cognition). Also interesting re LLM chain-of-thought methods https://t.co/pbailVpNc9 https://t.co/gQOclA4Kcl
A lot has been written about how the traditional science model of publishing and peer review is broken. One connection I havenโt seen made is with the **sensemaking** literature, which studies how people understand complex topics ๐งต 1/ https://t.co/Vr3dK4XkKK
Super excited to be returning to @allen_ai this summer for an internship with Kyle Richardson and the Aristo team!
Blue-sky proposal paper accepted for spotlight @ Bridging AI and Cognitive Science workshop, ICLR2020! https://t.co/i0ZKePIa4n Work with @GabiStanovsky @rtsarfaty and Dafna Shahaf (@HyadataLab) 1/n
Our position paper on the importance of metaphor and mental simulation in natural language understanding was accepted to #acl2020nlp! Metaphor is turning out to be quite a rabbit hole... ;) Crazy and fun ride @ChenShani2 @HyadataLab https://t.co/G8er6wyWpl
Our camera ready (pre-preprint) for #acl2020nlp is up! We propose a unifying view of embodied cognitive linguistics theory (drawing on @GeorgeLakoff & co.) with current NLU and embodied AI research, emphasizing role of mental simulation and metaphor https://t.co/mL9wUmduLy (1/n)
Looks like a cool idea - emergent grounded communication as a non-linguistic pretraining for NMT models. From Yaoyiran Li, @PontiEdoardo , Ivan Vulic, @annalkorhonen https://t.co/CEluPILXng
Some people imagine the future of science as LLMs peer-reviewing other LLM written papers, and that future seems pretty bleak to me. This excellent paper details the hazards of over-automating science and challenges us to re-humanize the scientific endeavor. Thoughts in ๐งต1/6 https://t.co/jvso2fMLH3
The next scientific revolution will be *metascientific* - in how we do science. IMO this paper presents a compelling vision for a new kind of science - more democratic, inclusive and relevant. ๐งต of related thoughts/excerpts > 1/ https://t.co/rZu2EmSwAf
@alienelf Totally, it feels like a massively funded distraction campaign from real problems (ones that big tech can't solve, as opposed to generating cool AI art). https://t.co/vfEMw6jj14
@ACMHT @daostack @InferenceActive @CSenseMakers @HyadataLab 2/TL;DR: lots of talk about free speech online, but what about *free attention*, and why is it key for sensemaking? Free speech=freely creating new content; open-sourcing/decentralizing code, finance, social media.. freeing attention=freely sharing *reactions* (likes, bookmarks,) https://t.co/tJVWCnPDi5
There may be a similar issue in NLP, where SOTA transformer models fail on synthetic language compositionality tests (e.g. https://t.co/sw9JJr0Q6k), and yet such "toy" tasks are often frowned upon. https://t.co/o68VZF1m5z
Loving this idea. Learning to "out-cooperate the competition" will spark massive and transformative systems change. https://t.co/G3U4ocTz97
I like how this @MetaAI paper opens with information overload as the main motivation but then does not include important critical perspectives on the risks of LLM based approaches to this problem, such as Situating Search by @chirag_shah and @emilymbender https://t.co/yc5SNy4ELT https://t.co/pll2VGXTYM https://t.co/TVnob0dOJg
Loving @aka_MatsuLab 's talk at @descieth @ CU Denver today! Moving from monoculture science to permaculture science with distributed & collaborative tools for thought๐ฑ๐ฑ๐ฑ https://t.co/IcY2xqPdfV
@gdb Alternatively, a big takeaway from the GPT paradigm is that our conceptions of what it means to be human are far more impoverished than almost anyone anticipated (e.g. thinking both humans and LLMs are merely some kind of stochastic parrots) https://t.co/CKj5CbWH5N
Active inference models of science ๐ https://t.co/HqqIizm5WV
Language models != world models "Probing Multimodal LLMs as World Models for Driving" https://t.co/EnOfNem3a6 https://t.co/kMAehfm8UU
๐ฅฑ Tired: Organizing the world's information (Google mission statement) โกWired: Self-organizing the world's information (decentralized stigmergic sensemaking systems) https://t.co/ArImnNJYM0
๐ง We need new frames for rethinking our relation to data! In existing frames, data is inert & detached, remains we leave behind that are meaningless to us (data as "exhaust" or "oil") but can be captured by corporations and converted to gold. ๐งต 1/ https://t.co/jyMB5GHkRB
AI was revolutionized by an old idea (neural nets) revisited with new tools (massive compute/data) Similarly, CI (collective intelligence) will be revolutionized by an old idea (semantic web) revisited with new tools (AI + decentralized tech) What makes the CI revolution harder >
@RadxChange #RxCDenver2021 was๐คฏ Our session on web3/ed-tech/sense-making turned into a brilliant 1-hr improv. idea jam (as many of the RxC sessions do). @RadxChange is an amazing hopeful force and I'm super grateful+happy to be able to join the effort! https://t.co/APcdPKqApy
๐งโ๐ฌ๐งชWe'll be presenting our work on process-level annotation+modelling of scientific procedural texts this week@ #EACL2021! Poster session 23/4 13-15 UTC, we'd love to chat :) Project page: https://t.co/jGNH2kIWKt Great team effort w/ @loadingfan, @alan_ritter, @GabiStanovsky! > https://t.co/cY8vI70nFQ
1/Check out our 1st @gitcoin grant at https://t.co/QsmA0N3x3c : a joint project of @CSenseMakers, @braddegraf, @lovevolv As we explored here, open sharing of attention signals (aka markers/annotations), eg bookmarks, tags, likes, is key to healthier information environments๐งต https://t.co/O0RmCBcDla
Very interesting work, obj.-oriented reps. + sequential application of learned rules - arch. biased towards both (1) learning of specialized modules as well as (2) their reuse. Compare w/ @robert_csordas https://t.co/JhcXKkD5O6 where Transformers found to succeed at 1 but not 2 https://t.co/RYIf8l19em
@ACMHT @daostack @InferenceActive @CSenseMakers @HyadataLab @HumaneTech_ @tristanharris @jbakcoleman 7/Stigmergy refers to the phenomena of indirect coordination mediated by modifications of the environment, eg ant pheromone trails. The crazy part is that the env acts as a kind of distributed memory system for a collective organism: https://t.co/HOrnJNF217 https://t.co/s289YsYM94
Interested in compositional generalization? Check out my write-up and (independent) implementation of @keysers et. al's DBCA, a great method for creating dataset splits with precise control over compositionality, to better evaluate comp. generalization. https://t.co/0SvdujsaGw
@pfau Also known as the "problem problem" https://t.co/XyWeUOJYnU https://t.co/F6muiGTRW7
@matspike lol a different kind of AGI singularity
Someone in the neighborhood yoga studio gifted me this book! Gotta love Cambridge ๐ค https://t.co/5AdZfiUjSz
So many of our problems with technology will be solved when more of the people creating tech will be the kind that like humans more than they like machines...
Cool Nature paper at intersection of neuro-cogsci:AI- learn to predict cortical responses to words based on word2vec decoding using a simple linear layer(!?), find language comprehension system widespread across both hemis, not localized in left hemi. 1/2 https://t.co/CAskJbEKuc
Cool paper from @ErikJones313 and @JacobSteinhardt drawing on insights from cog sci to analyze Codex code generation failures! While Codex papers too often just focus on models doing "magic tricks", this kind of error analysis is v. helpful to get a sense of model limitations https://t.co/BjaXyoKQHW
@yoavgo I don't think there ever will be any "good definition" becuase there are infinitely many ways sentences can be similar or dissimilar, where goals/task demands dictate which metrics will be useful in a particular context https://t.co/wKts2y1huW https://t.co/PGx2O4s13n
Semantic web is a killer app for web3 https://t.co/7k6HF12FQC
https://t.co/u8JM49OhA5 https://t.co/fvmDpEM5uB
What do ant trails have to do with navigating info online, and what can ๐๐ teach us about designing social networks for collective intelligence? Come explore with us at our talk + discussion at #DWebCamp2023! @braddegraf @CSenseMakers https://t.co/aWJl1ncLVl @GetDWeb @sched
I just donated to the Metacrisis on @gitcoin. Join me in making a difference by donating today! https://t.co/R7NA3Q5yxj
Language models != World models Inconvenient truths for AI... Excellent post by @mark_riedl https://t.co/tvBejydnmI https://t.co/CHijlDGu7g
Just remarkable how easily you can swap "cell" with "company" and get a coherent analogy between capitalism and cancer https://t.co/5VfZt1W07N https://t.co/ke7pmjdU45
1/Really enjoyed this talk - an insightful political philosophy perspective on how can we better govern collective attention in the age of global scale digital platforms (Facebook, Twitter, etc) dominating the public sphere. https://t.co/kADlmFP5Wd
Our collective brain is so broken. โneurons that fire together, wire togetherโ is a fundamental rule of learning in individual brains But in our *collective brain*, the firings (~individualsโ activity on various social networks) and wirings (~our social + interest graphs) > 1/ https://t.co/Uw2PRiqCGw
Really enjoyed this paper by @dohmatobelvis @introspe @danilobzdok , helps weave together different accounts under a coherent framework, and understand default mode network (DMN) role in "dark" language understanding. A particular idea that (1/2) https://t.co/R47wwrIaAA
@ACMHT @daostack @InferenceActive @CSenseMakers @HyadataLab @HumaneTech_ @tristanharris @jbakcoleman @philipplenz6 @mbauwens @project_solid @Filecoin @RubenVerborgh @mmasnick @protocollabs @obsdmd @memexgarden @logseq @AthensResearch 16/ interoperable protocols/storage for stigmergic primitives can enable bootstrapping existing PKP growth for *collective* sensemaking/knowledge. https://t.co/rEU1til8tF
This paper from Firston, Tenenbaum et. al looks ๐ฅ Active inference models *select their own data*- provides an interesting perspective on the age old q of RL vs supervised. https://t.co/UMDiiSC2Cy
Still amazing to be able to just hop into a deep (while also super friendly and informal) discussion with Jay McClelland and so many other brilliant minds. - Systematicity is highly educational rather than innate - Compositionality in language emerges from compositional envs https://t.co/kX9UOQmcOl
Finding @NotionHQ to be a fantastic tool for research (and free for students, too!) Two features that would make it a killer app for reference management: - Special web clipper for academic papers to extract citation metadata automatically - Ability to annotate embedded pdfs
This is a sign to print and point to https://t.co/BQzbQtdSsI
periodic PSA: Sensemaking is embodied https://t.co/g4hOyvTdev https://t.co/3K9irCNrch
DDOS attack on humanity. Pascal saw this coming in the 17th century https://t.co/6WemK2uWsg https://t.co/IpYz0Bpgg6
@ACMHT @daostack @InferenceActive @CSenseMakers @HyadataLab @HumaneTech_ @tristanharris @jbakcoleman 8/Agents modify the env, which affects same/other agents, constituting a cybernetic feedback loop. Pretty much all large scale coordination must rely on stigmergy (including for humans), so we really want to create healthy and well functioning stigmergic infrastructure! https://t.co/ORXbYJtH8x
@ACMHT @daostack @InferenceActive @CSenseMakers @HyadataLab @HumaneTech_ @tristanharris @jbakcoleman @philipplenz6 @mbauwens @project_solid @Filecoin @RubenVerborgh @mmasnick @protocollabs @obsdmd @memexgarden @logseq @AthensResearch @WilliamsJames_ @metagov @RadxChange 20/And thank you for your attention if you made it all the way here :) This is just the start of our journey and we'd be happy to connect with fellow sensemakers! To be in touch, follow us at @CSenseMakers or drop us a DM to join our collective! /end
@ACMHT @daostack @InferenceActive @CSenseMakers @HyadataLab @HumaneTech_ @tristanharris @jbakcoleman @philipplenz6 11/Markers are in many ways digital traces of human attention! So what's our proposal? Stated simply, akin to the open source software (OSS) movement to free software, we propose an Open Source Attention (OSA) movement to free stigmergic marker data (~attention) from platforms >
@yoavgo https://t.co/p3p8JVczUR
โwhat about downloading our lives to the land instead of just uploading them to the cloud?โ visited @awarenesss on the farm with @shirandudy , we loved it! Thanks @DoomerOptimism for connecting us via the podcast โจ https://t.co/hGHSgi2IvP
Intriguing connections with divergent hemispheric reasoning strategies - does over-reliance on the model-biased left hemisphere lead to proxy failures? @BrettPAndersen https://t.co/axzff8XTkj https://t.co/7UyLSoPJco https://t.co/IXKGshvCP2
Really looking forward to engaging with the local DeSci community @DeSciBoston ! Come by if you're around Oct 21-22 โจ๐งโ๐ฌ https://t.co/BebJlmEUJT
@ACMHT @daostack @InferenceActive @CSenseMakers @HyadataLab @HumaneTech_ @tristanharris @jbakcoleman @philipplenz6 @mbauwens @project_solid @Filecoin @RubenVerborgh @mmasnick @protocollabs @obsdmd @memexgarden @logseq @AthensResearch 18/As @WilliamsJames_ writes eloquently in https://t.co/Pvmm1zlIde, surmounting the formidable challenges faced today by humanity requires that โwe give the right sort of attention to the right sort of thingsโ. We hope our work can help galvanize action towards this vital cause!
@yoavgo ืืกืืื, ืืื ืืกืฃ, ืืืืช ื*ืื* ืคืฉืืื ืืื ืฉืฉืืื-ืืง ืื ืืืืฆืจ ืืจืื ืืขืืืช ืืกืงืืื ืืกืจ ืชืงืืื - ad-tech ืืืืจ, facetune + ืืืืื ืืืฃ ืฉื ื ืขืจืืช ืฆืขืืจืืช, metaverse, ืฆ'ืืืืืื ืืืกืืืคืืื ืืืจืฉืืื ืืจืืื https://t.co/QAK3hUkXpe
@ACMHT @daostack @InferenceActive @CSenseMakers @HyadataLab 3/Sense-making refers to processes by which agents make sense of their environment, by organizing sense data until the environment is understood well enough to enable reasonable decisions. But collective sensemaking is in a bad state... https://t.co/8XL7BU6iqb https://t.co/mHzMS1oFWs
@ACMHT @daostack @InferenceActive @CSenseMakers @HyadataLab 5/opaque information ecologies, where network data and algorithms are effectively hidden away from public/scientific/ethical oversight. Opaque platforms are involved in a host of deeply problematic social phenomena, as noted by many @HumaneTech_ @tristanharris @jbakcoleman https://t.co/helohDCwEU
The AI accelerationists get one thing right - AI *is* accelerating something - what it's accelerating is the realization that our systems are rotten. (education -> LLM essays graded by LLM "teachers", relationships -> LLM romantic partners) We can do so much better... https://t.co/9VMFa2EsAe
Thoughts after reading this interesting piece- 1) Replacing "AGI" with "Perfect Industrial Capitalism" magically makes x-risk debates a lot more coherent > https://t.co/TTSEtWbwvh https://t.co/LfSFU8EwNK
Some updates on Sensemaking Networks! (blog version https://t.co/un9BhoaDbz) We built our first prototype for an LLM-augmented semantic cross-poster that can nanopublish @nanopub_org alongside posting to social media! LLMs suggest keywords + the nanopub semantic structure 1/ ๐งต https://t.co/8BB0lVjWmD
Can we create this future please? "Collective Intelligence as a great scientific, technical, and political project that aims to 'make people smarter with computers, instead of trying to make computers smarter than people'" https://t.co/DXGwhKMe7j
@couldrynick @jakusg @sbenthall @csageland @doctorow Shout out to the Social Sensemaking Scenius where we're working and playing with these ideas! ๐ฑ โจ @CSenseMakers @relationalorg @lovevolv @InferenceActive @popp0x @gabriel_export @tysullberg @nebo333 @ShaRefOh @joyqzhang @getbloop @maceagon @clinamenic 13/13
More compelling evidence for why we need to collectively govern our own attention https://t.co/WyUA65b68i
reason #651 why it's time to rethink and rebuild science social media as a piece of critical scientific infrastructure https://t.co/ECG69roJcr https://t.co/nIgXoO1eSk
@jacobmenick @hllo_wrld Agreed! This is the ML analog of extended cognition- "the mind (model) does not exclusively reside in the brain (ANN) or even the body, but extends into the physical world (dataset)" :) https://t.co/71HzBooCqk
@cognazor "Ask not what's inside your head, but what's your head inside of" https://t.co/7vxpTi4Mu5
The classic โInformation wants to be freeโ (Stewart Brand, 1984) needs updating in our age of exponential information explosion. Access to information โ ability to *make sense* of information Next level: โInformation wants to be FAIRโ ๐ https://t.co/bkaIKYYr5B