I learn on the fly ✈️ 💻 • Co-founded @missio_ai • AI Engineer
Coming soon in your home town
Joined June 2021
- Tweets615
- Following994
- Followers234
- Likes1.7K
I had this insight:
Matrix Multiplication <=> Computing ALL combination of Dot Products
Q × K^T computes ALL 9 possible dot products between 3 queries and 3 keys
Here is an illustration I made of that fact
claude.ai/public/artifacts/5…
SlyRacoon retweeted
🥉 3rd Place (Subjective Track): DynastAI
model trained to efficiently play a story-based RPG, biased towards relevant pathways and events
@SRacoon23
After the @NousResearch Hackathon, it’s clear: AI engineers’ future lies not in complex prompts, but in crafting rich environments ( e.g terminal, browser, ide, etc)
SlyRacoon retweeted
Nous Research's RL Environments Hackathon recap thread!
Starting with the stars of the show, the winners!
Top 3 for the subjective track were:
1st - Pokemon Trainer - by @iyajainfinity & @AlexReibman
2nd - VR-CLImax by @JakeABoggs
3rd - DynastAI by David van Vliet and @SRacoon23
Top 3 for the objective track were:
1st - CyberMaxxing by @1999_karthik
2nd - HelpfulDoctors by @tsadpbb, Nilesh Shah, Max Phelps, and Alexander Speicher
3rd - Physical RL by @nullref0 and @venkatacrc
Another special shout out to our partners, @xai, @MistralAI, @nvidia, @tensorstax, @akashnet, @nebiusai, @runpod, @daytonaio, @morph_labs, @LambdaAPI and @Tesla
As well as our many judges from @arcee_ai, @axolotl_ai, @cursor_ai, @latentspacepod, @MIT, @togethercompute, @haizelabs, @SophontAI, @EdgeAGI, @Google, specifically:
@AlpayAriyak, @winglian, Samuel Barry, @tmm1, @keirp1, @swyx, @teknium, @karan4d, Meghana Puvvadi, @arattml, @brianlechthaler, Josh May, Alex Gu, @gordic_aleksa, @AlpayAriyak, @eraqian, @LukePiette, Rohan Rao, @chargoddard, @LoganGrasby, @xennygrimmato_, @zhangir_azerbay, @rogershijin, @max_paperclips, @theemozilla, and Abhinav Balasubramanian
🏆 Took 3rd place at the Nous Research Hackathon!
We built DynastAI — a medieval kingdom RL environment with adaptive rewards that evolve with agent choices.
PR: github.com/NousResearch/atro…
Demo: screen.studio/share/3GJ8riH5
@NousResearch @SHACK15sf
🚀 Meet Missio: Your Ultimate AI Teammate!
Project chaos? Overwhelming tasks? Let me handle it. I connect the dots across Slack, GitHub, Linear, and more to ensure nothing slips through the cracks.
producthunt.com/posts/missio…
Help me make waves on Product Hunt! 💥
👉 Upvote Missio
🚀 Just wrapped up the @ycombinator x @supabase Hackathon! 🎉
Introducing SupaTask: Why manage your to-do list when AI can do it for you? 🤖
Automated tasks, smarter productivity—built in a weekend! 💡
Check it out: [link to project or demo]
#SupaTask #AI #Productivity #Supabase
SlyRacoon retweeted
In their latest research update, our Interpretability team discuss their take on the next five hurdles between us and a mechanistic understanding of neural networks: anthropic.com/research/circu…
Early-stage offer: If you're a developer company with 500+ Discord users & 20+ daily support issues, I'll help for free this August. 5 years industry & 2 years freelance tech community experience (blockchain to AI). I'll personally tackle your support challenges.
DM for details!
This quoted post is unavailable.
🖋️ AI-Assisted Writing: Exploring Llama 3.1 8B
Just published a deep dive into Meta AI's Llama 3.1 8B model for poetry analysis and generation!
Curious about AI's role in creative writing? Read here: open.substack.com/pub/slyrac…
SlyRacoon retweeted
12 yo noticed that successive squares are separated by successive odd numbers (i.e. n^2 = (n - 1)^2 + 2n - 1) and was more excited by this than anything he's ever learned in class. I told him this is what real math is like.
Deployed my first LLama3 endpoint on @replicate
Don't have a GPU so I needed to deploy someones model on huggingface.
Glad it worked
SlyRacoon retweeted
1/6 1st prize of $ 5K for the Web3 native event platform on #NEAR bounty goes to NearEventHub, a platform of event management using blockchain to ensure commitment and facilitate trustless transactions among all parties. Congrats @krisos, @lyons_800 and @slyracoon23.
Online forums struggle with fragmented conversations. "Sliding window" threading uses AI to incrementally analyze messages and build dialogue flows.
We apply large language models (LLMs) powered by transformers within each window. The LLM assigns "logits" - perplexity scores indicating a message's probability of fitting different thread contexts.
This logits-based classification allows nuanced threading keyed to topical and contextual cohesion. The window continually slides forward for efficient, dynamic analysis as conversations rapidly evolve.
LLM threading enhanced by logits strategy significantly improves discussion quality and user experience compared to rule-based approaches. It handles complex digital interactions with multiple simultaneous topics.
We're interested in further optimizing compute. Does the community have experience whether KV-caching may help unlock more real-time threading.
🌟 Urgent Support Needed for Nicholas 🌟
Nicholas, a Ukrainian refugee, lost his car in an accident. It was his lifeline.
🎥 Watch his story: youtube.com/watch?v=f7AgIR7I…
🙏 Donate/Share: gofundme.com/manage/support-…
Even a RT can make a big difference! #SupportNicholas #HelpingHands
I am looking for 3 websummit tickets for @missio_ai
DM me if you are selling or if you have someone who does👇
