Connecter les entreprises avec l'IA & le Web3. ๐
Dubai
Joined June 2022
- Tweets1.6K
- Following97
- Followers17
- Likes2.7K
Peter van Dijk retweeted
Public Art Abu Dhabi Biennial: Artists, locations, dates and what to know thenationalnews.com/arts-culโฆ
Peter van Dijk retweeted
#News | Dubai Police Explore AIโs Role in the Future of Protection and Emergency Response
Details:
dubaipolice.gov.ae/app/home/โฆ
#ZeroBureaucracy
#DiscussionSession
#FutureForesight
Peter van Dijk retweeted
Peter van Dijk retweeted
Multi-agent work gets messy when every specialist lives in a separate chat.
AgentCrew is an open-source multi-agent chat application for builders handling coding, research, and automation tasks.
It helps you coordinate a team of focused agents by giving each one a role and tools, then letting agents hand work to the right specialist when a task needs another perspective.
Key features:
โข Specialist agent setup โ define agents with their own descriptions, tools, and system prompts
โข Agent handoffs โ use @AgentName to pass a subtask to another agent with its own tools and instructions
โข Multiple work modes โ run it through a desktop GUI, terminal, one-shot jobs, or an HTTP API
โข Provider choice โ connect subscription-based options, API-key providers, or custom local providers
โข Tool controls โ configure capabilities such as code analysis, file editing, web search, browser automation, and MCP tools
Itโs open-source (Apache License 2.0).
Link in the reply ๐
Peter van Dijk retweeted
Viggle's JST-2 takes one image and gives you an animation-ready character (a Gaussian splat) your agent can call, no rigging or skinning step. if you're vibe-coding a game, that's characters sorted.
๐ค Paid partnership
Today we are releasing JST-2, the world's first full-stack physics-native world foundation model.
Many world models learn in pixel/video space. We went back to first principles and built the entire model stack in 3D.
Generate a 3D scene in seconds and expand it in real time. Create characters with motion built in, all through one general architecture.
Peter van Dijk retweeted
OpenAI co-founder Andrej Karpathy just delivered the best 30-minute masterclass on Agentic Engineering: from vibe coding to Software 3.0.
00:44 - why programmers feel behind
02:28 - Software 3.0 explained
09:41 - why AI needs verification
15:46 - vibe coding vs agentic engineering
25:17 - agents everywhere
Prompts โ Vibe Coding โ Agents โ Engineering
most people will stop at generating code and call it progress
he explains why building reliable systems requires an entirely different discipline
you can outsource the thinking, but not the understanding
watch it today
follow me for more AI breakthroughs, engineering insights, and tutorials worth your time.
Peter van Dijk retweeted
holy f*ck, SpaceXAI engineers put together 54 Grok bots and every one is free
Shub Gaur, Lauren Tan, Matt Palmer and the rest of the team are behind them
the full list from the Grok Bot marketplace - each bot with a one-line breakdown of what it does
bots like these sell for $50-200 on other platforms
you go to the marketplace, pick what you need and hit install
the 72-hour build, the config files and the prompt engineering - all gone
how to use the video:
> let the highlight run through all 54
> pause on the one that covers the job you hate
> install that bot first
full marketplace below ๐
Peter van Dijk retweeted
The 50-day countdown to the Abu Dhabi Grand Prix at Yas Marina Circuit has begun!
#EmiratesNews #DubaiOneTv #UAE #Dubai #News
Peter van Dijk retweeted
This keeps popping up in my feed.
I think Iโm going to watch it even though itโs going to be depressing. Has anyone else seen it yet? Thoughts?
In 2014, Canadian and American salaries were roughly on par.
Today, American salaries are 30-50% higher.
Hard working Canadians are struggling to make ends meet and many families spend more on taxes than on housing, food, and bills combined.
'Broken Country' is a new documentary that explores what went wrong in Canada, what can be done, and what other countries can learn from it.
Watch it here: brokencountry.com
Peter van Dijk retweeted
Whatโs Inside an Iceberg Table? ๐ง ๐ฆ
An Apache Iceberg table looks like a regular table when you query it, but beneath the surfaceโฆwell, itโs like an iceberg. ๐
Underneath, youโll find a small tree of files: one JSON file at the top, a few layers of Avro files in the middle, and Parquet files at the bottom. When something is slow or looks wrong, it helps to know whatโs in that tree.
DuckDB can read all three formats, so you can inspect every layer with plain SQL. In this post, we create a small Iceberg table, change it a few times, and then walk through the files from the top down: duckdb.org/2026/10/13/insideโฆ
#ApacheIceberg #DuckDB
Peter van Dijk retweeted
๐จ Govt (DoT) Has Made Caller ID Mandatory Across All 4G and 5G Network From November 1
Incoming Calls on Your Phone Will Show Caller's Name, No More Unknown Numbers Calling
This Will Be a Nightmare For Spam Callers
Excellent Step From Department of Telecom
Peter van Dijk retweeted
ElevenLabs has an open-weights problem.
Fish Audio released Drama 3, a text-to-speech model whose weights you can download and run on your own hardware. It reportedly beats ElevenLabs on pronunciation accuracy.
It speaks 83 languages, Russian included.
The editing caught my attention most. You can regenerate a single word inside finished audio and leave everything around it untouched, so one mispronounced name no longer means re-rendering the whole clip.
here's what else it does ๐
โ switches voices in the middle of a sentence
โ generates dialogue between multiple characters in one pass
โ takes emotion, whispering, intonation, pace and even crying as plain-text directions
โ adds pauses and delivery cues to your script automatically before generating
That last one means you can skip learning a tagging syntax. You write the line and the model writes its own stage directions.
Running it locally is not a laptop job: the weights need 48 GB of VRAM. Without that, Fish Audio has a browser version to test it first.
Peter van Dijk retweeted
Dubai Police actually added a 920-horsepower Lamborghini Temerario hybrid supercar to their luxury tourist patrol fleet.
#EmiratesNews #DubaiOneTv #UAE #Dubai #News
Peter van Dijk retweeted
Google DeepMind has released a new GitHub repo called 'cryptids', ahead of the Gemini 4 Argon general launch later today.
They hold tiny Turing machines that run forever unless a famous conjecture is false, plus Lean proofs to do the same.
Google seems to be back :)
Peter van Dijk retweeted
Latest Updates:
- Sheikh Mansour bin Zayed has issued a resolution establishing the Community-Managed Virtual Authority, appointing Zayed bin Mohamed bin Zayed as Chairman.
- Global Village is set to open its doors tonight for Season 31.
#EmiratesNews #DubaiOneTv #UAE
Peter van Dijk retweeted
Is context engineering just a new name for RAG?
Not quite. But they're solving the same problem: building the right context for your LLM.
Here's how we got from one to the other โ and why it matters for AI data scientists.
Peter van Dijk retweeted
Agentic AI is not just about the model.
Itโs about the entire system around the model.
A production-ready enterprise AI agent typically needs 5 layers working together:
โ ๐๐ป๐๐ฒ๐ฟ๐ณ๐ฎ๐ฐ๐ฒ ๐๐ฎ๐๐ฒ๐ฟ
Chat, voice, APIs, webhooks, WebSockets and user-facing experiences.
โ ๐ข๐ฟ๐ฐ๐ต๐ฒ๐๐๐ฟ๐ฎ๐๐ถ๐ผ๐ป ๐๐ฎ๐๐ฒ๐ฟ
Workflow engines, task routing, planning, state management and multi-agent coordination.
โ ๐๐๐ ๐๐ฎ๐๐ฒ๐ฟ
Model routing, prompt management, memory and reasoning.
โ ๐๐ฎ๐๐ฎ ๐๐ฎ๐๐ฒ๐ฟ
Vector databases, embeddings, document processing, knowledge graphs and retrieval.
โ ๐๐ป๐ณ๐ฟ๐ฎ๐๐๐ฟ๐๐ฐ๐๐๐ฟ๐ฒ ๐๐ฎ๐๐ฒ๐ฟ
Compute, deployment, monitoring, security, compliance, CI/CD and governance.
And this is where things get interesting.
Choosing the โbestโ LLM is only one piece of the puzzle.
The real enterprise challenge is making ๐ฎ๐น๐น ๐ณ๐ถ๐๐ฒ ๐น๐ฎ๐๐ฒ๐ฟ๐ work together reliably, securely and at scale.
Because an AI agent that can reason but cannot access the right data is limited.
An agent with great data but poor orchestration is unreliable.
And an agent that works in a demo but fails in production isn't really enterprise-ready.
The future of Agentic AI will be built as a ๐๐๐๐๐ฒ๐บ, not just selected as a model.
Which layer do you think is the most challenging to build well?
โค๏ธ Like
๐ Retweet
๐ Bookmark
Peter van Dijk retweeted
Here are 10 new free learning resources from Harvard, Google and Karpathy, even if you've never studied AI.
Peter van Dijk retweeted
This is one of the things that I really really think has been true no matter how you do memory
Usually it's a combination of <obsidian-like> and <git like> primitives
Peter van Dijk retweeted
One ready blueprint, zero code, and youโve got a hedge-fund-grade AI trader running while you sleep
most people donโt even try, thinking itโs PhD-level. in reality AI does the hard part, the human part is simple
the trick nobody talks about: one model is too slow to trade, the other too dumb to think. so you run both
Opus 5.5 thinks and writes the strategy. Jev reacts in milliseconds. hard code between them holds the risk so the account canโt blow up
full architecture and complete Rust code in the paper below
this isnโt a one-scroll read. save it and build it properly when you have time
