Beyond the tech world? There is nothing. 🌐
London
Joined June 2023
- Tweets6.4K
- Following220
- Followers59
- Likes18.2K
AIKeaton UK retweeted
On a recent episode of @Heavybit’s Generationship, Vasanth Mohan (@v_mohan_) of @SambaNovaAI joined host @RachelChalmers to discuss what developers contribute once #AI agents can write the code. They cover engineering judgment, training junior developers, and what developer relations look like when documentation must serve both humans and machines.
Listen to the full episode below:
Thanks for having @v_mohan_ on for an episode of Generationship! @rachelchalmers @heavybit
Loved this conversation on the future of dev relations.
AIKeaton UK retweeted
Microsoft was due to open a datacentre in Essex.
They pulled out.
Why?
Because of “power supply concerns”.
Guess who’s on the board? Nick Clegg himself.
Nick Clegg said that nuclear power “wasn’t even an option” because it wouldn't be open until the 2020s.
Good thing we don’t need it now.
Small businesses, industries, and households that are struggling because of these choices.
They refused to build nuclear power then, and still make it extremely difficult to build today.
Take Hinkley Point C. It was supposed to open in 2017, almost a decade ago.
Now? It MIGHT open in 2030. If it’s on time.
To make matters even worse, we are letting an unelected quango delay and delay construction.
Natural England is imposing a £700 MILLION “fish deterrent system”. It is predicted to save a grand total of one salmon every twelve years.
Even then, they might add on more costs.
Households are suffering, and we are letting this odd (unelected) organisation HURT them even MORE.
It’s quite obvious that we have a problem.
It’s time to ABOLISH “Natural” England, and END the delays to building nuclear power.
What do you think?
AIKeaton UK retweeted
If you are wondering whether your website is affected, we have packaged a Codex skill to help check.
Point your Codex at this repo to guide you through it: github.com/openai/php2shell-…
OpenAI models helped researchers—operating as part of our Patch the Planet initiative—discover critical security flaws in PHP, an open-source programming language used to run an estimated 70% of websites.
Over the past two months, we led a coalition of maintainers, researchers, and security providers to prepare fixes and protections ahead of disclosure. This duration was necessary because the fix was quite intricate and required a major change to the PHP engine. Every site using PHP can now safely upgrade their version to stay secure, in addition to applying mitigations from various security vendors.
Due to the severity of these flaws and PHP’s widespread use, we consider this bug to be one of the most impactful vulnerabilities in recent memory. It also marks a new phase, where AI can find security bugs with internet-scale impact.
This is an example of the Defender’s Window in action: to our knowledge, frontier AI was able to find this vulnerability ahead of attackers, giving defenders time to prepare.
openai.com/index/coordinated…
Readers added context they thought people might want to know
OpenAI may claim a role in the coordinated response and patching process. But the initial discovery and exploit development, however, occurred several months earlier by an independent researcher.
x.com/lcfr_hax/statu…
AIKeaton UK retweeted
We're open-sourcing EdenLive together with EdenGolem v0.1, an agent that turns a single prompt into a playable world with rules and state running underneath.
EdenGolem builds the state. EdenLive brings it to life.
Explore EdenLive:
-> Project: vast-eden.ai/?p=edenlive
-> Code: github.com/VAST-AI-Research/…
-> Model weights: huggingface.co/VAST-AI/EdenL…
-> Technical report: assets.vast-eden.ai/assets/e…
AIKeaton UK retweeted
The opposite of token-maxxing isn’t giving up on AI. It’s using the right model for the right task. The new Auto router in @NotionHQ handles that for you.
AIKeaton UK retweeted
It’s great how none of this is mysterious:
1) buy lots of SMT machines
2) load with the most popular parts
3) steer customers to use those parts with surge pricing for less common stuff that would require humans swapping reels
totally possible in the US
AIKeaton UK retweeted
Notion, Stripe, Netlify and a handful of companies are pre-AI era companies are really benefiting from the rise of agents
How? Through docs that are human and agent friendly
Through welcoming agents happily via dedicated entry points:
> netlify.ai/
> projects.dev/
Through re architecting core services and adding robust MCPs
It’s not accidental, it’s because of hundreds of little decisions like the one here in this video
I've been personal agent-maxing lately, and loving it. The new products are proactive, useful, and so fun. They get even better when they know me well.
So I built a portable context vault in Notion and connected every agent to it. They write back new memories, skills, and preferences, then sync each morning.
Now, whenever I try a new agent, the first thing I do is connect my vault, and it instantly gets all my preferences in one shot. I can also start a project with one agent, and switch over to another seamlessly.
I shot a quick video to walk through the setup. LMK if you have any questions!
AIKeaton UK retweeted
You must have seen this video. A tile glows under every step, the crowd cheers, and then our own SM champions start saying: "Japan is years ahead. What is India doing?"
Look at what the clip can show and what it cannot.
It can show the flash. Pavegen's own FAQ says one step gives about 3 joules, a 5W bulb for one second. A 60-second clip can fit dozens of flashes. It cannot show the daily total, the cost, or whether the tiles are still there next year.
The daily total exists, because Japan measured it.
Tokyo Station, 2008. JR East laid 90 sq m of piezo mat on stairs and at a ticket gate. Target: 500 kilowatt-seconds a day. It worked.
Weekdays beat the target, and the peak day gave 766 kWs. That is about 0.2 kWh, a 100W bulb for 2 hours.
So what worked: the physics, and the target.
What did not: scale. I found no report of a permanent rollout.
Washington DC, 2017. Pavegen's own estimate for three strips of tiles was 1.25 kWh a day. A local paper called the output "vanishingly small".
The data page promised to the public had a broken footstep counter. Pavegen itself says one step is about 3 joules, a 5W bulb for one second.
India tried too. In 2016 a Nagpur engineer, Mandar Tulankar, put an energy floor at Platform 1 of Nagpur station. It lit four 25W bulbs overnight. He said it captured only about 30% of its potential. His company is listed as closed on Crunchbase. I found no report of RDSO clearance / a later installation.
Footsteps do make power. Not enough to matter yet.
Sources: JR East release 11 Jan 2008; Takefuji, Keio Univ. (NOLTA 2008); Washington City Paper 23 Mar 2017; Pavegen FAQ; Crunchbase.
AIKeaton UK retweeted
Forrester has named Google a Leader in the @forrester Wave for Public Cloud Platforms. + Google received the top score in the Current Offerings Category!
Read more and download the complimentary report → goo.gle/4jcekF1
AIKeaton UK retweeted
BYD just launched a luxury sedan with over 500 miles of range for 32k.
Charges in 9 minutes.
You can’t buy them or anything even close in the States.
I don’t like this feeling.
AIKeaton UK retweeted
Slow growth, higher taxes, and a deteriorating business environment in the UK aren’t ideal for family offices, but one part of the sector remains strong: non-resident, non-domiciled individuals and families setting up family office branches in London.
Article | famcap.com/2026/10/non-dom-n…
AIKeaton UK retweeted
“Babe, wake up, new “What If” piece just dropped:
What if Britain had never broken up the electricity board?
In 1990 it split the CEGB into four companies and sold them. Now it’s buying the pieces back one acronym at a time: NESO in 2024, GB Energy in 2025, GB Grid in 2026.
Meanwhile, about £1.5bn went on switching off wind the wires couldn’t carry.
AIKeaton UK retweeted
We care a lot about design and craft at @interfere_. Rather than just say it, we decided to show our work and publish our design system. Brought to you by @jakubkrehel and I :)
design.interfere.com
AIKeaton UK retweeted
The room is overflowing for @corentinanjuna and @dzbarsky’s talk on HermeticBuild at BazelCon 26
AIKeaton UK retweeted
MS shares its picks in 4 buckets.
Compute: $NVDA $TSM $AVGO
Networking: $GLW $LITE $COHR Soitec
Memory: $SKHY Samsung Kioxia
Enablers: Aspeed, Murata, $ASX
Morgan Stanley: Memory Channel Checks
HBM / Nvidia Rubin
→ HBM ASPs expected +121% YoY in 2027. HBM4 pricing at $3–5/Gb, with HBM4e at $5+/Gb.
→ Nvidia Rubin ramp slower than planned. Fewer than 10k VR200 NVL72 racks expected in 2026 vs. an original ~15k target. Rubin expected to become mainstream after 1H27.
→ HBM4 yields remain constrained. 8-high stacks offer ~5 percentage points higher yields than 12-high and are expected to become mainstream in 2027. Lower HBM capacity per chip could benefit RDIMM/SOCAMM as inference KV-cache shifts to host memory.
DRAM
→ 4Q26 contract pricing revised to +10–15% QoQ from +3–8%. PC DRAM +13–18%, server DRAM +10–15%, mobile DRAM +0–5%.
→ US hyperscalers reportedly placed additional RDIMM orders for 2027. Korean suppliers' PC DRAM output could fall >20% YoY as capacity shifts toward server DRAM and HBM.
→ DDR5 modules now trade at a ~10% discount to DDR4. DDR4 supply expected to decline ~20% in 2027 as suppliers exit legacy production.
→ DRAM capex projected at $98bn in 2026 (+82% YoY) and $150–160bn in 2027 (+~60%), but meaningful additional bit supply only arrives in 2H27.
NAND / Enterprise SSD
→ 4Q26 NAND pricing revised to +15–20% QoQ from +0–5%. Enterprise SSD pricing expected +23–28%, significantly outpacing client SSD.
→ Several major US cloud customers reportedly shifting from TLC SSD + HDD to QLC SSD for AI inference KV-cache storage. QLC costs roughly one-third as much as TLC at equivalent capacity.
→ Consumer NAND inventories are building despite enterprise SSD shortages, with suppliers reportedly bundling excess consumer inventory to secure sales.
AIKeaton UK retweeted
Atlas is starting to look like a real automotive worker（Currently undergoing internal verification and testing）.
xitter.cf/CyberRobooo/status/210…
Atlas is starting to look like a real automotive worker.
Boston Dynamics has opened its RMAC(Robotics Metaplant Application Center)inside Hyundai Motor Group’s Metaplant America in Georgia, giving Atlas a real-world environment to train and validate manufacturing tasks.
Engineers can teleoperate Atlas with a VR headset, teaching it tasks such as automotive parts logistics, sequencing, and preparing components for assembly.
The plan is to put Atlas into HMGMA production operations in 2028, starting with parts sequencing, then expand into more complex component assembly by 2030.
Earlier,Hyundai Motor Group says it plans to deploy 25,000 Atlas robots across Hyundai and Kia’s global plants over the next few years, while building a U.S. facility capable of producing 30,000 robots per year.
AIKeaton UK retweeted
Modal just passed $1 BILLION in ARR, up from about $300m in May
Modal is a serverless cloud where developers run AI inference, training and batch jobs on GPUs without managing any infrastructure
The company just raised $750M in a Series D led by Accel at a $15.75 billion valuation, more than triple the $4.65 billion it was valued at four months ago
The market does not fully understand the demand for compute
