Joel Miller @joelmiller
AI consultant. Co-founder @ExoBrainAI... exploring AI and reducing the capability versus adoption gap! exobrain.co.uk London Joined March 2008-
Tweets206
-
Followers365
-
Following548
-
Likes42
The ExoBrain newsletter this week looks at mobile/ambient AI, AWS and the competing cloud hyperscalers, and mixed signals from the markets... xobrn.co/54kethnc
The ExoBrain newsletter this week looks at whether AI can resuscitate global healthcare, unpacks next-gen Llama 3, and ‘delves’ into LLM linguistics... exobrain.co.uk/weekly-ai-news…
Introducing Meta Llama 3: the most capable openly available LLM to date. Today we’re releasing 8B & 70B models that deliver on new capabilities such as improved reasoning and set a new state-of-the-art for models of their sizes. Today's release includes the first two Llama 3…
The ExoBrain newsletter this week explores music generation service Udio, the AI adoption versus opportunity gap, and a wave of new model announcements… exobrain.co.uk/weekly-ai-news…
The future of attention just dropped, and it looks a lot like a state space model (finite size, continual updates) Little doubt now that a mixture of architectures will support the long-term, gradually conditioned memory needed for highly capable agents Buckle up!
The future of attention just dropped, and it looks a lot like a state space model (finite size, continual updates) Little doubt now that a mixture of architectures will support the long-term, gradually conditioned memory needed for highly capable agents Buckle up!
New paper out today in @NatMachIntell, where we show robust neural to speech decoding across 48 patients. nature.com/articles/s4225…
Some more details from @EricSteinb about @magiclabs (see why @karpathy @danielgross @natfriedman @polynoamial all invested) - LONG context - more than TENS OF MILLIONS tokens support - Inference time compute idea - Agents with 99.9% accuracy - AGI within a decade thanks…
Some more details from @EricSteinb about @magiclabs (see why @karpathy @danielgross @natfriedman @polynoamial all invested) - LONG context - more than TENS OF MILLIONS tokens support - Inference time compute idea - Agents with 99.9% accuracy - AGI within a decade thanks… https://t.co/N9SucIqXr4
The @ExoBrainAI newsletter this week looks at quantum AI, Google’s disrupted business model, and what ‘AI safety’ really means… exobrain.co.uk/weekly-ai-news…
Even @AnthropicAI are "speechless" in response to some of the things Claude 3 can come up with!
Even @AnthropicAI are "speechless" in response to some of the things Claude 3 can come up with!
Based on some pretty conservative estimates, the raw/efficiency power of Blackwell is essentially going to make it the workhorse for the intelligence revolution...
Based on some pretty conservative estimates, the raw/efficiency power of Blackwell is essentially going to make it the workhorse for the intelligence revolution... https://t.co/2RESfP50yr
I can imagine a future Hollywood where banks of highly trained visual 'mentats' think movies into existence...
I can imagine a future Hollywood where banks of highly trained visual 'mentats' think movies into existence...
Quite a thought... a full @nvidia 'AI Factory' (32k Blackwell GPUs) would probably be able to train GPT-4 in under 3 days. 🫨 Down from the ~100 days it took on A100s. Imagine when multiple companies are churning out several GPT-4 class models a week???
The ExoBrain #ai newsletter this week explores how large language models are increasingly breaking free from their chatbot constraints and accelerating autonomous work, robotics, and community software creation. exobrain.co.uk/weekly-ai-news…
Amazing progress being made on fine tuning models with consumer GPUs... This post includes some great explanations around the key techniques: answer.ai/posts/2024-03-…
8 years ago today, AlphaGo beat Lee Sedol in a milestone for AI. Unlike typical neural nets, AlphaGo spent ~1 minute per move improving its policy via search. This boosted its Elo by more than a 1000x bigger model. Even today, nobody has trained a raw NN that is superhuman in Go.
The ExoBrain newsletter this week covers the amazing new capabilities of Claude 3 from Anthropic. We also analyse cost and capability across the mainstream AI model landscape, assessing all the major new entrants… xobrn.co/7dakryex
Wow more 'sparks'; meta-awareness!
A REALLY good take on the implications of long context LLMs and RAG by @jerryjliu0. Plus a reinforcement of @llama_index's mission on building production grade data infrastructure for all LLM applications (not just RAG) 💪🦙🔥 llamaindex.ai/blog/towards-l…