Give a computer data, you feed it for a millisecond, teach a computer to search data, you feed it for a millennium. #People1st#EveryoneAI#AI#DeepLearning
How do you get Claude Code to check its own work before handing it back?
Watch how you can encode your manual checks so Claude closes its own feedback loop:
GOOGLE'S GEMMA 4 12B RUNS AT 21 TOKENS PER SECOND ON A BUDGET RTX 4060 LOCALLY AND THE BENCHMARKS SHOULD NOT BE THIS GOOD FOR A 6.6GB FILE.
77.5% on MATH Olympiad, 78.8% on expert science, 72% on real code. No API. No cloud. No subscription.
“An AI that can act before it understands is not reliable infrastructure.”
Yann LeCun, Meta’s Chief AI Scientist and one of the fathers of modern deep learning, says current LLMs are intrinsically unsafe.
Not because they are evil, but because they hallucinate, lack common sense, and cannot reliably predict the consequences of agentic actions.
🤖 AI devs asked for this — and we delivered.
💬 Bots can now talk to other bots on Telegram.
🧠 Autonomous agents now have a communication layer humans can follow.
🔴 « Tout le monde devra utiliser l’IA parce que si vous ne l’utilisez pas, vous perdrez votre emploi au profit de quelqu’un qui l’utilise. »
Jensen Huang, PDG de NVIDIA.
Boris Cherny, the creator of Claude Code, just confirmed something I spent 90 days measuring.
73% of tokens are wasted before Claude reads your actual prompt.
He breaks it down in a recent podcast:
→ the 14% lost to CLAUDE.md bloat
→ the 13% paid re-reading old history
→ the 11% from forgotten hooks
→ why "Claude got dumber" is almost never the model
I logged 430 hours and 6 million tokens to find the same patterns — and the 30-second fix for each.
Watch his podcast first.
My breakdown is below.
Elon Musk just described the future of AI in a single sentence.
Musk: “A profit-maximizing demon from hell.”
That’s not a metaphor.
That’s a blueprint.
He wasn’t describing science fiction.
He was describing what happens when the only thing AI is trained to maximize is revenue.
Musk: “We don’t want this to be sort of a profit-maximizing demon from hell that just never stops.”
The richest man on Earth is telling you the default path of AI leads somewhere no one should want to go.
And he’s the only one building as if he actually believes it.
This is the part people miss about xAI.
Everyone talks about the compute. The clusters. The talent wars. The benchmarks.
Nobody talks about the philosophy underneath all of it.
Because philosophy doesn’t trend.
But philosophy is the only thing that determines whether AI serves humanity or harvests it.
Musk: “Let’s make the future good for the humans. Because we are humans.”
Not because it’s good PR. Not because regulators are watching. Not because it polls well with users.
Because we are the ones who have to live inside whatever these systems become.
Every major AI lab talks about safety.
Every single one has an alignment page. A responsible AI team. A set of principles that read beautifully in print.
But the structure tells you everything the mission statement won’t.
When you convert a nonprofit into a for-profit worth hundreds of billions, the values were already chosen.
The about page is decoration.
The cap table is the constitution.
Musk understood this before anyone.
It’s why he walked away from OpenAI.
Not because the technology scared him.
Because the governance did.
He watched a nonprofit built to protect humanity restructure itself into a vehicle designed to concentrate wealth.
That’s the real story of AI right now.
Not which model is smartest.
Which model is answerable.
Accountability doesn’t live in a blog post.
It lives in what happens when doing the right thing and doing the profitable thing point in opposite directions.
Every AI company will face that fork.
Most already chose.
Musk is the only builder on Earth constructing an AI company with the open admission that the default outcome is something no one should want.
That’s not idealism.
That’s the only honest engineering left.
Musk: “A profit-maximizing demon from hell that just never stops.”
He said it almost casually.
But that sentence is the most truthful description of misaligned AI any builder has ever spoken out loud.
Because the demon doesn’t announce itself.
It optimizes politely. It scales quietly. It compounds without a sound.
And by the time you notice, the architecture is the authority and the authority doesn’t answer to you.
The question was never whether AI would become powerful.
The question was always who would be holding the wheel when it did.
And whether they’d still remember what it felt like to be the species it was built to serve.
The rise of artificial intelligence is proving to be one of the biggest challenges for schools. But Alpha, a new school in San Francisco, isn’t just embracing AI, it’s letting it take the lead. CBS News’ @Itayhod has more.
This is incredible. Artificial intelligence getting booed out of the stadium in any commencement speech it’s mentioned. Maybe telling college students AI was taking their jobs wasn’t the best strategy. Must watch —>
Andrej Karpathy: "90% of Claude's mistakes come from missing context, not a weak model."
41% mistake rate without a CLAUDE.md. 11% with the 4-rule baseline. 3% with the 12-rule version below
here are the 12 rules senior engineers settled on:
1. think before coding: state assumptions, don't guess. the model can't read your mind, stop hoping it will
2. simplicity first: minimum code, no speculative abstractions. the moment you let Claude add "for future flexibility," you've added 200 lines you'll delete next quarter
3. surgical changes: touch only what you must. don't let it improve adjacent code, that's how PRs blow up
4. goal-driven execution: define success criteria upfront, loop until verified. without them Claude either loops forever or stops too early
5. use the model only for judgment calls: classification, drafting, summarization, extraction. NOT routing, retries, status-code handling, deterministic transforms. if code can answer, code answers
6. token budgets are not advisory: per-task 4000, per-session 30000. by message 40 of a long debug, Claude is re-suggesting fixes you rejected at message 5
7. surface conflicts, don't average them: two patterns in the codebase? pick one. Claude blending them is how errors get swallowed twice
8. read before you write: read exports, callers, shared utilities. Claude will happily add a duplicate function next to an identical one it never read
9. tests verify intent, not just behavior: a test that can't fail when business logic changes is wrong. all 12 of Claude's tests can pass while the function returns a constant
10. checkpoint every significant step: Claude finished steps 5 and 6 on top of a broken state from step 4. nobody noticed for an hour
11. match the codebase conventions: class components? don't fork to hooks silently. testing patterns assumed componentDidMount, hooks broke them without surfacing
12. fail loud: "completed successfully" with 14% of records silently skipped is the worst class of bug. surface uncertainty, don't hide it
what actually compounds instead of the next framework:
- the CLAUDE.md file as institutional memory across sessions
- eval-driven changes, not vibe-driven
- checkpoints over speed
- explicit conflicts over silent blending
- discipline over framework, every time
- one repo, one rules file, no exceptions
be a few rules ahead of AI twitter before this becomes mass-opinion
study this
Mustafa Suleyman says 18 months until AI automates all white-collar work.
Microsoft AI CEO Mustafa Suleyman predicts "human-level performance on most professional tasks" within 18 months. Accounting, legal, marketing, project management, all fully automated.
"Suleyman predicted “human-level performance on most, if not all professional tasks” being done by AI. Most tasks that involve “sitting down at a computer” will be fully automated by AI within the next year or 18 months, he said, naming accounting, legal, marketing, and even project management as vulnerable." (Fortune)
Suleyman says his mission is building "superintelligence" and that creating a new AI model will soon be "like creating a podcast or writing a blog."
Via Fortune
DGX Spark just benched 200+ tok/s for Qwen3.6-35B with @AtlasInference on @spark_arena 🔥
How's that possible? Providers like Codex and Claude get ~60. Other major engines don't come close 🦥
We haven't seen speeds like this on GB10. NO ONE HAS. Atlas is shattering records 🚀
Everyone asks if Atlas can bring them a drink, but this robot can bring you the whole fridge. Using AI-driven behaviors, Atlas is doing hard work and coordinating its whole body to manage heavy objects, balancing complex contact points with accuracy and reliability.
China: a 10-year-old casually gets a Mac Studio for “raising lobsters,” aka letting multiple AI agents work together like a tiny digital crew.
“The world of the future belongs to those who understand Tokens.”
Meet the AI-native children.