Pinned Tweet
My TED talk has now had over 1.5 million views
(the power of Ted rather than the power of Tom)
thul.me/28WL2sR
This "messy middle" is now hitting everyday services too, quietly breaking legal claims, insurance, ombudsmen, and customer support.
For decades, organisations relied on human effort as a natural rate-limiter. Writing a formal legal claim, regulatory complaint, or detailed appeal took hours of work or hundreds of pounds in legal fees. That friction kept inbound volume matched to the humans sitting on the other side of the desk reading them.
Generative AI just dropped the cost of creating those claims to zero.
Today, anyone can generate a 20-page legal claim or dispute letter in thirty seconds. Look at the UK Legal Ombudsman's latest figures: more than half (51%) of complaint files now show clear signs of AI drafting, helping drive a 37% surge in annual caseloads.
Because most receiving organisations still rely on humans to read and triage every submission, zero-friction AI generation acts like a live DDoS attack on their teams.
You cannot solve a machine-speed flood by hiring more people to read PDFs. The only way through this messy middle is AI customer service on the receiving side too, triaging, verifying, and resolving issues at machine speed. That is a big part of why I am so excited about our investment in @SierraPlatform (@btaylor and @claybavor).
Tom Hulme retweeted
Flint from @Recursive_SI is now on OpenRouter!
Recursive's first model: an open-weight 40B MoE (~2.5B active) built for coding agents by their autoresearch system, the Eureka Machine. 262K context, reasoning and tool calling.
Recursive reports 54.9% on Terminal-Bench 2.1, 89.7% on SWE-bench Pro v2 and 74.2% on SWE-bench Verified.
Free to use at launch.
Try it now: openrouter.ai/recursive/flin…
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.
Tom Hulme retweeted
We’ve developed the world’s most complete megakernel framework. It can perform entire training runs, serve inference, play video games, and even do all three simultaneously in a single kernel.
Tom Hulme retweeted
My bar for AGI is still an AI winning a Nobel.
But an AI researcher looking at the lm_head (that *everyone* tells you to keep in high precision) deciding to quantize it to 4 bits AND being actually right (0.7pt from bf16) is definitely impressive.
Even better when the result are open-weights pushed on the Hugging Gace hub.
Congrats @Recursive_SI !
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.
Tom Hulme retweeted
The @recursive_si team just introduced Flint, a state-of-the-art local coding model. Flint is 2x faster than comparable models and was built in under four months — an impressive pace from an equally impressive team. Congrats @RichardSocher and the Recursive team!
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.
Evolution created the 20W human brain through open-ended search, pushing compute to the edge so latency wouldn't let entropy win.
@Recursive_SI is running accelerated evolution. Instead of hiring more researchers, their Eureka Machine built Flint in under 4 months.
wired.com/story/recursive-se…
Tom Hulme retweeted
Great summary of our work on Flint by @cong_ml
1/ Introducing Flint, our first model at @Recursive_SI !!
A blazing-fast 40B model on the local coding Pareto frontier, shaped by autoresearch 🔥
Built from scratch in under four months with our Eureka Machine and an incredible team.
Here's how 🧵
Inference moves to the edge ->
A capable coding model that runs fast on a laptop, with your code never needing to leave your machine...
Accelerating progress from the team @Recursive_SI - give Flint a try...
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.
Being British in the AI era is a thermodynamic liability.
I want to be default courteous so I don't train myself to bark orders like a sociopath—but every standalone "Thanks, that's great!" forces an H100 cluster to reload 20,000 tokens just to tell me "You're very welcome!"
AI might be ruining my "mental VO2 max."
Instead of 60 mins of uncomfortable, high-lactate deep thinking (Zone 4), I now spend all day managing 8 smart digital agents—doing Zone 1 editorial cruising broken up by 10s context-switching sprints.
High output. Zero hypertrophy.
Tom Hulme retweeted
Thanks @JensenHuang & @nvidia for supporting UK startups. We started @Recursive_SI in both London and SF from day 0 because of the outstanding talent and ecosystem here.
Tom Hulme retweeted
Gemini 4 Argon is a step change in model performance.
In recent weeks, it has had a transformative impact on the Science projects in our team at @GoogleDeepMind and our external collaborators, and helped make a number of remarkable advances across scientific disciplines. 🧵👇
Another Easter egg added, this time with Gemini4 -->
My @GeminiApp Agent and I have rebuilt my personal site, and included an Easter 🥚, will be interesting to see if anyone finds it -->
thulme.com
Incredible to be 90% voice in the interaction 🗣️
Tom Hulme retweeted
#1 customer question since 2017:
“When can I send an avatar to a meeting for me?”
Today, we’re launching @synthesiaIO Sessions - live Avatar conversations that help scale your time, so you can:
- Train 1,000 sales reps in a day
- Do 1:1’s with 2,000 employees at once
- Interview 3,000 customers a week
Try it free 👇
Tom Hulme retweeted
Headphone (1) Pro launches today, and V of BTS is joining Nothing as our new Global Brand Ambassador. Here are a few moments from our shoot in Seoul with an artist whose instinct for fearless experimentation is exactly what we built this product for.
Tom Hulme retweeted
At our new startup, there are ZERO non-competes and we give each employee the choice of either 1 or 3 months notice periods, upon their request.
Competition is for winners.
BTW, today is the last day to join the founding team.
1/ Today we’re launching an open letter calling on the government to end unfair restrictions that stop Britain’s best switching jobs or starting companies.
UK AI companies that have raised $5bn agree: our best should be free to start and scale. 🧵
@downingstreet @biztradegovuk
