George Adamopoulos retweeted
Paper: openreview.net/pdf?id=lY7miv…
Credit to @Wurst_Imperium for the awesome meme pic! It's on the poster, on the spotlight slide, and on my t-shirt :)
George Adamopoulos retweeted
Got a spotlight at the ICML mechinterp workshop tomorrow!
MultiSTEVE-1s - a model zoo and interp suite of murderous Minecraft derps, ahem, agents.
Come say hi at poster spot #317 in Hall C at 11:00. Spotlight talks at 10:30.
George Adamopoulos retweeted
I've been working on a @3blue1brown style video explaining our research on attention patterns of OpenAI's Minecraft agent VPT.
1min teaser below.
You can see a longer WIP version in my recent talk at @cohere (from 24:55):
youtube.com/watch?v=p_vwKfhX…
George Adamopoulos retweeted
Announcing the AI Futures Project: an AGI forecasting and governance organization led by @DKokotajlo. Our first project is AI 2027: a scenario forecast of the development and effects of superintelligence. We've also developed an AGI tabletop exercise.
George Adamopoulos retweeted
The scenario: ai-2027.com
PDF version: ai-2027.com/scenario.pdf
Looking forward to engaging with comments!
George Adamopoulos retweeted
"How, exactly, could AI take over by 2027?"
Introducing AI 2027: a deeply-researched scenario forecast I wrote alongside @slatestarcodex, @eli_lifland, and @thlarsen
Why does advanced voice mode on the chatgpt desktop app use ~20% of my PC's RTX 3060? Is it just to visualize the blue flowing circle animation?
543
George Adamopoulos retweeted
Sorry, villagers! A good reminder that training agents to emulate humans has dark downsides that need to be addressed before empowering agents in any situation of consequence. Also, seems like very interesting work!
Breaking: @OpenAI’s Minecraft agent VPT kills villagers!
In other news: our paper on dissecting VPT with various interpretability techniques was accepted at the Mechanistic Interpretability workshop at ICML.
1/7
George Adamopoulos retweeted
Don’t miss our work at the Mechanistic Interpretability Workshop at @icmlconf on July 27 at 11 a.m. CET! While I won’t be there in person, @Karolis_Ram, @GeorgeAdams101, and @soniajoseph_ will present it. 5/5
#ICML2024
George Adamopoulos retweeted
Work done with the amazing team: @GeorgeAdams101 @MehrabHamidi78 @steph_milani @M_R_Samsami @artemZholus @soniajoseph_ @tyrell_turing @irinarish @ozgur_simsek_
From: @UniofBath @Mila_Quebec @mcgillu @UMontreal @polymtl @cerc_aai @CarnegieMellon
6/7
George Adamopoulos retweeted
Breaking: @OpenAI’s Minecraft agent VPT kills villagers!
In other news: our paper on dissecting VPT with various interpretability techniques was accepted at the Mechanistic Interpretability workshop at ICML.
1/7
George Adamopoulos retweeted
Join our 7th Scaling workshop shorturl.at/FQ0DM organized by @cerc_aai and @cerebras on Mon 7/22 in Vienna (Mariott across the street from ICML) - and our Open-Source AI Party on Wed 7/24, by @cerc_aai @cerebras & @togethercompute!
teapaw.org/CerebrasSystems/status…
teapaw.org/CerebrasSystems/status…
Registration is open for our next #ScalingLawsWorkshop next week.
📅 Monday July, 22
📍in person : Vienna (#ICML)
📍 Remotely : Via Zoom
🗒️ Free (but mandatory) Registration : shorturl.at/FQ0DM
#CERCAAI
George Adamopoulos retweeted
Interested in seamlessly updating your #LLM on new datasets to avoid wasting previous efforts & compute, all while maintaining performance on past data? Excited to present Simple and Scalable Strategies to Continually Pre-train Large Language Models! 🧵arxiv.org/abs/2403.08763 1/N
George Adamopoulos retweeted
Proud to announce the first open source time-series foundation model Llama (inspired by @Meta's LLaMA - thanks, @ylecun ;) , built jointly with my awesome collaborators from @cerc_aai @UMontreal @ServiceNowRSRCH and @MorganStanley: @arjunashok37 @krasul, @CluelessAndrew, @hena_ghonia, @rishika2110, @Arian_Khorasani, @653mjd, @GeorgeAdams101, Roland Riachi, Nadhir Hassen, Marin Bilos, Sahil Garg, Anderson Schneider, @NicolasChapados, @alexandredrouin, @vzantedesc, Yuriy Nevmyvaka.
George Adamopoulos retweeted
Foundation models are well-established in vision and language, but time series forecasting has lagged behind - it still relies on dataset-specific models.
Meet Lag-Llama: the first open-source foundation model for time series forecasting!
George Adamopoulos retweeted
sub.thursdai.news/p/dec-28?r…
📅 Last @thursdai_pod episode of this year, covered tons of new open source from @upstageai @unum_cloud @NousResearch @ldjconfirmed , the AI lawsuit, the ml-ferret models from Apple, and the Robin suite of models from @cerc_aai , we also had a round of predictions!
Also the fairly controversial quality of Midjourney v6 output (especially on Starwars) and collected tons of predictions (next tweet) for the 2024 year of AI!
( we also had a full interview w /@irinarish @kshitijkgupta @GeorgeAdams101 about Robin suite which will be posted as it's own episode soon)
George Adamopoulos retweeted
Replying to @irinarish @cercaai @UMontreal @kshitijkgupta @DanielZKaplan @simonramstedt @MistralAI @teknium
Credits: this repo and the associated fine-tuned models were built in collaboration w/ @simonramstedt Alexis Roger, @DanielZKaplan @CluelessAndrew Edwin Fennell @GeorgeAdams101 @kshitijkgupta Prateek Humane @QuentinAnthon15 @rishika2110 Sun Qi, Yuchen Lu, @irinarish (PI).

























