
Scientist @AmazonScience | Prev: @brave, @Samsung AI, @Cambridge_CL, @CERN | Coffee drinker, traveller, tech/car enthusiast | Opinions are my own
London, England
Joined December 2011
- Tweets554
- Following693
- Followers495
- Likes1K
Looking f/w to giving a talk at Imperial College London on Tue 17/11.
One Model, Many Budgets: Towards Adaptive Foundation Models
We'll discuss how we can move beyond one-size-fits-all inference and towards models that adapt their computation to the task and available resources
3
10
411
CS grads are oftentimes hardwired into thinking worst-case instead of average or happy-path scenarios. It’s a bias that stems from cont. problem solving, Big-Os, robustness, fault-tolerance, etc
FWIW, I find the OG framing rather terrible, but useful mental exercise nonetheless.
why are all the smartest 18-23yo stanford/harvard/mit people such losers mindset about everything
discuss any idea and they'll by default talk in great detail what things might not work, instead figuring out how to make it work
1
99
So, reviewers complain about ICLR AI generated submissions, and then authors complain about AI generated reviews. And both come from the same pool of authors. 🫠 This encoder-decoder intermediary clearly has some self-improvement to do.
1
7
567
New AR EAGLE-4 Drafter: Bringing the auto-regressive speculative decoding up to speed. However, adding trademarks™ everywhere in the contributions is so weird ... What's the expectation here? 🤨
75
While a big proponent on-device AI, the dynamics aren't in favour of the switch yet, at least for general assistants. It currently remains too expensive and the diff in utility is not apparent to your avg consumer. Collaborative inference & computation onloading are first steps.
sam: local inference is a terrible allocation of scarce resources
“a lot of people want local inference, and somehow they assume that the problem is solved now because we have open-source models, we're going to be able to run them on our computers, and so we're not going to need all the stuff in the data centre
but from a global allocation perspective, actually it's quite terrible to have everybody, like, have a bunch of the world's limited supply of DRAM sitting in their own computer and being used five percent of the time
you want maximum utilisation in the cloud if you're trying to deal with this very supply-constrained environment
somehow many people's intuition is the opposite of that”
1
1
86
Also, in consumer home settings, energy efficiency and sourcing are very different to the datacenters. Last, you don't need one local AI capable device per user, but potentially per household. We have talked about such concepts here: arxiv.org/abs/2210.10514
1
24
With the wave of reverse engineering everything, I am looking forward to better printer software/firmware! It's been terrible for too long ... 🫠
3
302
I love the surge of tinkering with your purchased hardware and RE their S/W. It's been long overdue & gives control back to the user. Security through obscurity doesn't cut it. With covert surveillance on users, orgs risk their reputation. Banning tool usage, also reprehensible.
66
Why are people putting NeurIPS in their handles? It's still October ... Reminds me of Christmas decorations 🎄just after Halloween 🎃
1
1
186
Commendable effort! 👏
1. Still, it's a single point of failure, isn't it?
2. Also, 3.5 months is a sprint. Probably today's landscape resembles more of a marathon. Can such an effort scale in time? Maybe ...
We formally announced Mirendil less than 4 months ago. Today, I'm excited to share an important milestone for:
Pretraining at scale without a pretraining team.
Mirendil's mission is to solve Recursive Self-Improvement (RSI) and use it to democratize frontier AI development, ultimately accelerating science and technology. At Mirendil, when we think about RSI, we think holistically about the company as an intelligent entity that self-improves toward its goal. From that perspective, the most powerful notion of acceleration is reducing the time and resources (number of human experts, compute, and other costs) needed to achieve ambitious goals.
Given that pretraining is perhaps the most important yet most challenging part of frontier AI development, we decided to run an ambitious controlled experiment: Starting fully from scratch, can we significantly reduce the time, number of human experts, and compute resources needed to go all the way to successfully pretraining a strong coding model at scale?
Mirendil-Base is the result of this experiment. Our first pretrained model was built by one researcher in 3.5 months, with less than $10M in compute. This includes all the research and small-scale experiments it took to get there, as well as the final training run.
The recipe holds up as the models get bigger. For reference, Mirendil-Base is in the 100B-500B parameter range, Mirendil-Base-Large in the 500B-2T range, and Mirendil-Base-XL in the 2T-10T range.
We're beginning to see a world where a single researcher, empowered by AI, can take on work that once required an entire research organization. As these systems improve, the bottleneck shifts from the size of a team to the ambition and judgment of the people directing them. This has profound implications for who gets to build AI and, ultimately, who gets to push the frontiers of science and technology forward.
We’ll share independent evaluations, post-training benchmarks and API-access sooner than you think so stay tuned!
85
Cool, local and open! Can’t wait to test it. I am looking fw to the paradigm of local-to-remote handoff and vice versa. Not all tasks are equal, neither in terms of difficulty, nor in terms of deadline. Adaptive by design.
Introducing Flint, @Recursive_SI’s first model: a fast local coding model built with our automated research system, the Eureka Machine.
• Best-in-class fast local coding
• Up to 2x faster than comparable models: 125+ tokens/sec on a MacBook Pro, 150+ on RTX Spark
• Built from scratch in under 4 months
Private, unmetered intelligence on your own machine. Open weights available now on @huggingface, run it locally on your favorite framework or try it for free on @opencode or @pidotdev.
69
I read so many tweets today on Revolut’s outage and its repercussions. I get it, but having a single point of failure in your personal finances may not be a wise thing. Obviously no blame on users here, but anything can fail … Please diversify to mitigate such risks.
71
Model worms will not even need to break in and self-replicate. People are already giving them direct access. “While you’re in there …“ 🤠
1
107
Or as I would like to call ... Moving "tick" to "tock".
OpenAI, Anthropic and other leading AI labs face a growing challenge: Better models aren't necessarily attracting more usage. A 10-point intelligence advantage now earns just 20% more developer traffic, down from nearly 69% in late 2025.
Full story: thein.fo/4hGJ9jY
78











