AI researcher mapping human–LLM capability boundaries. Building validation systems and games to explore how humans and AI interact.
United States
- Tweets 37
- Following 4,054
- Followers 3,008
- Likes 28
Pinned Tweet
my website is done!😄👇
You are welcome to make suggestions.
aihumanbench.com
6
9
96
2,277
If human brain's long-term memory is ~2.5 PB:
• HDD: ~$60K
• Enterprise NVMe: ~$350K
• DDR5 ECC RAM: ~$80-100M
Storing 2.5 PB is cheap. Making it instant, associative, & computational is astronomical.
Good news: Your brain just appreciated in value.😀
3
8
66
899
⚡ How fast are you?
400+ ms: Refocus & retry
300–400 ms: Beginner
200–300 ms: Average
150–200 ms: Pro Gamer
120–150 ms: Elite
Under 100 ms: Superhuman? Probably a false start or prediction.
Test your 5-round average:
AIHumanBench.com
#ReactionTime #Gaming pmc.ncbi.nlm.nih.gov/article…
3
5
47
626
AI can already perform an enormous amount of computation.
But more compute doesn’t automatically mean better judgment.
Humans still have an important advantage in many real-world situations: knowing when something looks wrong, ambiguous, or uncertain.
3
2
34
539
Ever wondered how your brain stacks up against AI? 🧠⚡
Test your reaction speed, memory, logic, and problem-solving skills
Ready to find out? 👉 aihumanbench.com
#AI #HumanVsAI #BrainTest #AIHumanBench
1
19
297
Memory isn’t one ability.
Different tests can measure very different things:
• Short-term: brief retention
• Working: hold + manipulate
• Spatial: locations + relationships
• Auditory: sound patterns
A low score in one task doesn’t mean you have a “bad memory.” #AI #Agent
ALT An infographic distinguishing short-term, working, spatial, and auditory memory, with a reminder to compare results only between similar memory tests.
1
2
16
208
Your reaction-time score isn’t just reflexes.
Four things can change it:
• Sleep and vigilance
• Focus and distractions
• Practice and anticipation
• Display and input latency
For a fairer comparison: use the same device, run 5 attempts, and compare the median—not your best score.
1
9
139
Where do humans still outperform AI?
The answer changes with every model release.
AIHumanBench is mapping that boundary across reaction, memory, logic, creativity, and social reasoning.
We’ll publish the tests, results, and limitations—not just the conclusions.
ALT An abstract research map visualizing the changing boundary between human and AI capabilities across reaction, memory, logic, creativity, and social reasoning.
1
6
207
Explore the project and take the first benchmark:
aihumanbench.com/
Which human ability should we test first?
2
75
The first CPU built for agents is going to work at scale.
@SpaceX is deploying NVIDIA Vera to accelerate the orchestration, code execution, and data processing that powers its next generation of agentic AI — keeping GPUs fed and agents acting fast.
From gigawatt AI factories to orbit. One NVIDIA architecture, everywhere.
280
583
5,419
9,897,417
Nice to see modularity pushed this far — even the agent loop, sessions, storage, and UI are swappable. Curious to see how well it holds up in real-world use.👏
🧩 DeepSeek Harness v0.1 is now available in Developer Preview!
🔹 We’re opening it up to developers building agent harnesses worldwide and open-sourcing the codebase in MIT license.
🔹 Powered by the Cordis meta-framework, DeepSeek Harness is an agent harness built around one core idea: Everything is a plugin. Models, tools, skills, sessions, sandboxes, filesystems, loops, orchestration, and UI are ALL implemented as plugins, and can be mixed, matched, replaced, and extended.
Try it now!
github.com/deepseek-ai/deeps…
4
239
Just tried it—seriously impressive.
FLUX 3 Video is #2 in the world.
To celebrate, FLUX 3 Video is free to use in our playground until Sunday 16th, 11:59pm PT (link in the thread).
This is just the beginning of what’s coming. Up next: 4K, video editing, and multiple images & videos as reference input.
1
5
194
Exciting news: Qwen3.8-Max by @Alibaba_Qwen is #2 in Image-to-WebDev Arena!
With 1,631 pts, it’s trailing only Claude Opus 5 (Max) by 39 pts!
Congrats again to @Alibaba_Qwen on this huge release!
📢Meet Qwen3.8-Max — our most capable model to date.
Next week, the open weights of Qwen3.8-Max will be released, and Qwen3.8-27B is also going open-weights to meet you all!🎉
Qwen3.8-Max, a new bar for coding and cowork at 2.4T parameters:
- Autonomous coding: 10+ days of self-evolving development, from empty folder to production without hand-holding, complete project trace in the GitHub：github.com/qwen-code-dev-bot…
- Real work, real results: Production-quality deliverables across hundreds of professions.
- Long-horizon mastery: System-level autonomous planning with closed-loop adaptive learning, driving 500+ turns of chip design optimization and 365 days of e-commerce strategy.
- Native multimodal intelligence: Vision isn't just input — it's a continuous feedback loop for planning, execution, and self-correction.
💰Pricing:
Input: $2.0 / M tokens
Output: $6.0 / M tokens
Implicit Caching: $0.25 / M tokens
Start building with Qwen3.8-Max! 🚀
📖 Blog: qwen.ai/blog?id=qwen3.8
✅ Qwen Studio: chat.qwen.ai/?models=qwen3.8…
⚡ API: qwencloud.com/models/qwen3.8…
43
77
972
163,807
Does anyone else feel like Claude Code's usage limits are super generous now?
A few months ago with Opus 4.5, I'd hit my limit by day 3 every week. Now with Opus 5, I can't even max it out.
I maintain a 100k-line codebase and don't even need Fable 5. #codex #claudecode #ai
3
1
191