Liam "OD" retweeted
The Al Ain Book Festival just launched a podcast called Sawt Al Ain
Recording the stories of old Al Ain so the next generation can actually hear them
A lot of those stories were told out loud and never written down
Good to see someone finally pressing record
Liam "OD" retweeted
Let me share some details. This ranking is based on livability on a local salary, not a foreign one. I gave equal weight to quality of life (safety, healthcare, pollution, climate, commute) and to how far the median local paycheck goes once taxes, rent and living costs are paid. Essentially, it shows where locals and people working in a city live well.
To build it, I relied on official wage data for each city (the US Bureau of Labor Statistics, Eurostat and so on), applied local tax rules to that salary, and used Numbeo for rent outside the centre and daily costs. Just over 300 cities in total. There are some nuances: Gulf cities are out, for example, because there's no reliable median wage data there.
We all know Numbeo is relative, but we had to start somewhere...
Liam "OD" retweeted
SSMC in Abu Dhabi just launched a Women and Children Institute with 18 specialties under one roof
Every mother who has spent a week driving between clinics with a baby in the back seat just felt seen
Liam "OD" retweeted
Dubai Chambers has more than 85 thousand Indian owned businesses as active members
And it is working with Nasscom on agentic AI
Indian talent and Dubai ambition
This partnership keeps writing itself 🇦🇪🇮🇳
teapawt.org/EconomicTimes/status/2…
Dubai courts global IT companies, startups for AI push
economictimes.indiatimes.com…
Liam "OD" retweeted
Replying to @nateberkopec
Just use Pixi Mask. Your agent prophylactic.
github.com/Panoplos/pixi-mas…
Coding agents are part of every engineering team now, and Turnkey is giving them secure Git signing.
The infra we built for securing crypto private keys now supports agent signing for Git. Another step towards securing the guardrails for the agent economy.
Liam "OD" retweeted
6 Solana builder deadlines in the next 3 weeks.
▪️ Nov 26: Apply to the @dev3pack Global Hackathon. University students, solo or teams of up to 4, building Nov 27–29. $10K+ for Best App Overall, Seeker phones for the best mobile app. Submissions close Nov 29.
▪️ Nov 27, 21:00 ICT (14:00 UTC): Ship in @magicblock Solana Blitz v10. Online, opens today at 21:00 ICT. $1K USDC. Eligible projects must use Ephemeral Rollups or VRF.
▪️ Dec 3: Apply to The Bridge Accelerator. 8 weeks, free, 0% equity, for underrepresented founders. Dev3pack hackathon winners are fast-tracked.
▪️ Dec 4: Apply to Solana Incubator Cohort 6 (@incubator). 3 months in NYC from March 2027. Early applicants prioritized.
▪️ Dec 6, 04:00 ICT (21:00 UTC Dec 5): Submit to the Perps and Prediction Markets hackathon (@SolanaFndn). $100K main track.
▪️ Dec 9: Submit to the Busan Blockchain Hackathon, co-hosted by Solana Foundation and DB Securities (@SuperteamKorea). KRW 20M pool. Online qualifier, open to anyone. Demo day Dec 15 in Busan.
Bookmark this.
Liam "OD" retweeted
#Akvan crossed the N #AndhraPradesh & S #Odisha coast near #Gopalpur as a cyclonic storm between 730 & 830 IST with wind speeds of 70-80 km/h & gusts to 90 km/h. Rain will continue in these areas into Saturday. Scattered rains will return to #TamilNadu & #Chennai early next week.
Liam "OD" retweeted
Sometimes, when I switch to DeepSeek v4.1 Flash 1017 in my Pixi harness to work on some OSS stuff, I am forced to think that the labs are getting it all wrong with their model RL.
DS just gets shit done so quickly and so succinctly -- and so cheaply!!!!
I am going to start testing DS as the driver for the Software Factory and report on his successful this is, because Fable and Astra are driving me nuts with their constant distractions, attention to details where the value is little and overall slowness.
Liam "OD" retweeted
running a processor is easy to set up and easy to forget about, until something quietly goes wrong
that's why i like what the community built with Acurast.watch
it's an unofficial tool for people running processors on @Acurast and it surfaces the things you'd otherwise have to track by hand: app updates, slashing risks, battery status and more
slashing risk is the one that stands out to me
if a processor is out of date or running low on battery, you'd want to know before it costs you, not after
it also says something about the network that community members are building their own monitoring layer on top of it
that usually happens when people are running the thing seriously
if you operate processors, it's worth a look
@Acurast
Liam "OD" retweeted
DP World just took its investment in the Port of Maputo in Mozambique to 200 million dollars
Dubai has a habit
It finds a port somewhere in the world and makes it busier
Liam "OD" retweeted
Proud that Dubai has topped the Local Online Services Index (LOSI) in the UN E-Government Survey 2026, which assessed 194 cities worldwide. Six years ago, Dubai ranked 16th, but our ambition was to be number one. Our government entities worked as one team to make digital services faster, simpler and more accessible. Building on this achievement, we enter a new phase of harnessing AI to anticipate people’s needs and deliver services that exceed expectations. Our goal is for everyone living in Dubai to experience the benefits of this progress in their daily lives.
Liam "OD" retweeted
This advice is terrible if you are building complex systems with a lot of contextual awareness per domain.
We tried this approach for the first iteration of Genba and even Kobo (the Genba Software Factory), and the results were horrible.
This by Anthropic seems both right & a condemnation of many “agentic solutions” & harnesses being built at companies. Rather than a sprawl of specialized subagents built by different teams, you often get better & cheaper results using one agent to do it all claude.com/resources/article…
Getting two AI models to agree is easy when they are answering the same simple question. Getting reliable judgment when money and contractual obligations are involved is a different challenge.
That's one of the problems discussed by @kstellana with @HenriArslanian.
In an agentic economy, transactions won't always go as planned. Agents can interpret agreements differently, disagree over outcomes, and create disputes that need more than another automated response.
@GenLayer approaches this through validators running different AI models, which assess disputes and reach consensus on questions requiring judgment.
The bigger challenge is making that process trustworthy, not simply making AI faster.
What matters more when AI models resolve a dispute: agreement between models or the ability to verify their reasoning?
Liam "OD" retweeted
Magan and Dilmun
Our region had names long before it had borders
Now Zayed National Museum and Bahrain National Museum are digging together for six weeks at Qalat Al Bahrain
Neighbours going through the family archives
.@znmuae and Bahrain National Museum have launched The Magan–Dilmun Dialogues project, bringing together archaeologists and museum professionals from the UAE and Bahrain for a six-week excavation at Qal’at al-Bahrain, a UNESCO World Heritage Site.
This @primus_labs post reframes privacy as a speed problem, and AutoHoG gives it a number I can picture: a 64-bit transfer in 90 milliseconds, over 210 transfers a second on eight RTX 5090s. I like that no new hardware and no decrypted data are needed, because privacy stops being a tradeoff and becomes infrastructure institutions can run. The test is whether finance feels as fast as open finance. Which use case will institutions try first?
Privacy Was Too Slow To Use.
I've been looking at @primus_labs, and AutoHoG is the part that finally made FHE feel real. Not a new promise, but a speed number.
Traditional FHE evaluated one simple gate at a time. A 64-bit operation turned into thousands of gates, and thousands of bootstrappings. That was the wall.
AutoHoG designs custom gates instead, folding mergeable circuits into a single bootstrapping. A confidential 64-bit transfer now takes about 90 milliseconds.
On eight RTX 5090s, that is over 210 transfers a second. No special hardware. No decrypted data.
What makes it interesting is what speed unlocks. Privacy stops being a tradeoff and starts being infrastructure institutions can actually run.
The real test is whether confidential finance can feel as fast as open finance. At 90 milliseconds, it almost does.
Liam "OD" retweeted
Fun talk digging into the details of how we decoupled aspire from .NET while using .NET 😅
youtu.be/tT-rgopqhsU?is=O_QA…
Liam "OD" retweeted
I tried something on @Acurast that made the whole idea of phone-powered computing click for me.
A game of Tetris.
Sounds pretty ordinary, right? Except the decisions weren't being made by my computer.
They were coming from an actual Android phone connected to the Acurast network.
Here's what caught my attention.
Acurast's Laya demo generates five possible placements for each falling piece. An AI model running on a phone evaluates those options and chooses where the block should land.
When I tested it, the dashboard showed:
• Decision processed in 821 ms
• Average response time of 826 ms
• 11 decisions served by the connected device
• Computing device located in Spain
• 6 CPU cores and 7 GB RAM
The response came back in less than a second.
And there's something useful about being able to see which device actually handled the request, rather than treating the entire process like a black box.
Think about that for a moment.
A smartphone somewhere in Spain was handling an AI decision for a game I was accessing from another location.
No massive GPU setup needed for this particular task.
That's a practical example of what Acurast is building: a decentralized compute network where compatible smartphones can help developers run applications.
Of course, Tetris is only a demonstration. It doesn't prove every AI workload belongs on a phone.
But it shows why smaller, focused tasks could benefit from distributed computing.
Want to see it yourself?
Try Acurast's Laya Tetris demo, watch a few moves and check the device information behind each decision.
The interesting part isn't the game. It's discovering who's actually doing the computing.
🤝 Paid partnership
Near Ai Gateway app is live.
Created by me and @kpoldsam.
It's a super easy way to have run Near's TEE private AI models just a couple of clicks away in Claude, Codex and Open Code.
Near Ai enable you to have full sovereignty over your AI usage including what you do with your data.
You can choose to share your data with the Commons for discounts on your Ai usage.
Get it or have your agent install it from.
gateway.near.ai





























