So I tested DeepSeek V4.1 Flash on real tasks. 2 repos, 105 hidden bugs, find and fix what you can.
Opus 5 (max): 27
Grok 4.6 (max): 27
DeepSeek V4.1 Flash (max): 24
GPT-5.6 Luna (xhigh): 23
Opus 5 (high): 21
A really strong model for everyday tasks. And look at the cost:
Opus 5 (max): $51.33
Grok 4.6 (max): $16.96
DeepSeek V4.1 Flash (max): $1.80
GPT-5.6 Luna (xhigh): $2.50
Opus 5 (high): $38.77
It's at the Pareto frontier.
Other effort levels (high) and extra tests for DeepSeek V4 Pro dropping every hour in this thread 🧵
A 35B language model running on an iPhone using only 1–2.5 GB of peak memory.
No cloud. No remote server. No desktop GPU.
Today, we’re open-sourcing Edge0 — a framework for running large AI models fully on-device.
witnesses describe the Dario as visibly emotional during the briefing, reportedly telling officials that unrestricted open weights are a danger to humanity, and separately, in an entirely unrelated agenda item, that they are also a danger to the valuation narrative of an upcoming public offering.
the two concerns are not connected, sources emphasize.
Where does a humanoid robot pay for torque?
Every joint answers it differently, and so does every robot maker.
- Legs and hips absorb hard ground impact, so they favor low-gear planetary drives that survive shock.
- Wrists and arms need precision in a tight space, so they favor compact harmonic reducers.
- Hands: put motors on the joints, in the palm or the forearm
- @Tesla uses linear actuators, which push in a straight line, for high-force joints like the knee.
Each choice has different implication of the supply chain dependency.
Claude Code has 93 commands.
You only need these 16 to get 80% of the value 👇
/init
→ generates CLAUDE.md (your repo memory)
/plan
→ think first, code later
/context
→ shows what’s eating your tokens
/compact
→ free up context mid-session
/clear
→ reset between tasks
/model
→ switch Opus / Sonnet / Haiku
/btw
→ ask side questions without breaking flow
/rewind
→ undo when things go off track
/agents
→ run multiple sub-agents in parallel
/chrome
→ control browser (click, fill, navigate)
/loop
→ run prompts on repeat (monitoring)
/simplify
→ 3 agents review your code together
/permissions
→ pre-allow safe actions
--dangerously-skip-permissions
→ full auto mode (no interruptions)
Shift + Tab
→ switch modes instantly
"ultra think"
→ max reasoning depth
Most people use Claude Code like a chatbot.
That’s why they struggle.
Power users run it like a system.
Save this. You’ll need it. ♻️
#AI #Claude #AIAgents #LLM #GenAI #DevTools
This would be a good starting point for those who want to understand Large Language models and Artificial Intelligence in simple language - https://t.co/4KQCIbviyC
This guy quietly built the best physics textbook on the internet, and he gives it away for free.
His name is Bartosz Ciechanowski.
He is an ex-iOS engineer who writes interactive articles about physics, math, and engineering as a weekend hobby.
The site is called https://t.co/o0y0n63P48.
Open the article on the Mechanical Watch. Drag the watch to spin it. Pull a slider and the case peels away so you can look inside. Wind the mainspring by hand. Slow down time and watch the escapement click, the balance wheel breathe, the date wheel turn.
Nothing is a video. Everything is a live 3D simulation running in your browser tab.
He has done this for:
→ The Moon and its tides
→ GPS satellites
→ Gears and mechanical watches
→ The internal combustion engine
→ Bicycles and how they stay upright
→ Cameras, lenses, and light
→ Sound waves
→ Airfoils and how planes fly
Every article is thousands of words long with dozens of hand-coded interactive diagrams.
He writes them alone in his spare time. He publishes only a few per year. In his own words, he wants to keep the content accessible to everyone without ads or paywalls.
No publisher. No university. No company behind him.
One person is quietly shipping better science education than most universities charge tens of thousands a year for.
Most consumer health startups go after healthy consumers first.
@ThroneScience is starting with people who already live with gut conditions, where automatic tracking has an immediate use beyond curiosity.
I spent the day with co-founder @ScottHickle and the team to see how they’re bringing an unusual $18M-funded hardware product to market.
The route to making a smart toilet mainstream may start with the people who already need it most.
New INSIDE Startups episode out now.