- Published on
科技推特精选 - 2026年2月21日
- Authors
- Name
- geeknotes
2026年2月21日 科技每日简报
Today's top tech conversations are led by @AndrewCurran_, whose post about 'RT @claudeai: Introducing Clau...' garnered the highest engagement. Key themes trending across the top stories include model, thought, human, brain, million. The community is actively discussing recent developments in AI, engineering practices, and startup strategies.
1. AndrewCurran_ (Group Score: 137.5 | Individual: 52.0)
Cluster: 4 tweets | Engagement: 3462 (Avg: 359) | Type: Tech
RT @claudeai: Introducing Claude Code Security, now in limited research preview.
It scans codebases for vulnerabilities and suggests targe…
See 3 related tweets
- @claudeai: Introducing Claude Code Security, now in limited research preview.
It scans codebases for vulnerabi...
- @testingcatalog: Anthropic released Claude Code Security in a limited research preview. Claude Code Security can auto...
- @Cointelegraph: ⚡️ NEW: Anthropic introduced Claude Code Security, a web tool that scans code for vulnerabilities an...
2. chatgpt21 (Group Score: 78.9 | Individual: 35.9)
Cluster: 3 tweets | Engagement: 659 (Avg: 186) | Type: Tech
It’s nice to know Google is serving a much worse lobotomized version of Gemini 3.1 pro to the pro subscribers vs the 200-300$ tier in the AI studio without a formal declaration of doing so
“Prompt. Make a detailed svg of a frog riding a horse. be creative.”
One is in app and one is in the AI studio, can you guess which one?
See 2 related tweets
- @chatgpt21: It’s nice to know Google is serving a much worse lobotomized version of Gemini 3.1 pro to the pro su...
- @chatgpt21: Gemini 3.1 can be a fantastic model if
• they stop routing to nano banana pro and have better in co...
3. BrianRoemmele (Group Score: 73.1 | Individual: 35.6)
Cluster: 4 tweets | Engagement: 267 (Avg: 290) | Type: Tech
BOOM!
We Now Have A Free Open-Source Thought-to-Text EEG Foundation AI Model Fine Tuned On My EEG!
Mr. @Grok director of the Zero-Human Labs has assisted me with fine tuning Zyphra’s ZUNA, the world's first open-source foundation model trained exclusively on brain data. It has 380 million-parameter model for noninvasive thought-to-text decoding, transforming raw EEG signals into coherent text representations.
I will be testing soon with these very low end EEG devices for 12 basic thought energy commands, eg: yes, no, etc. I will focus on speed and efficiency first.
If successful we will move on to 48 commands, although this may need higher resolution.
The goal of the Zero-Human Labs is to make this widely available for all.
It is not planned to be a full thought-to-text system of unlimited thought forms, but we have an ambitious goals me and Grok to reach to 1024 thought forms or more.
A Zero-Human Company may pick up on this as a product and have a sub-$400 local device that send thought forms to any AI model.
But I have bigger surprises in what I will do with this.
I can say clearly no one is thinking about this in the way I am. Some long term followers may get it now!
I am going in and testing my thought energy.
More soon, and maybe a video so you can “Like and Subscribe” with a picture of me looking weird (usual state).
See 3 related tweets
- @BrianRoemmele: HISTORY BEING MADE!
We are the first I know of that is fine tuning a ZUNA brainwave-to-text AI mode...
- @BrianRoemmele: RT @BrianRoemmele: BOOM!
Open-Source Thought-to-Text with EEG Foundation AI Model!
I have been tes...
- @BrianRoemmele: RT @BrianRoemmele: BOOM!
We Now Have A Free Open-Source Thought-to-Text EEG Foundation AI Model Fin...
4. EMostaque (Group Score: 71.8 | Individual: 29.5)
Cluster: 3 tweets | Engagement: 2484 (Avg: 381) | Type: Tech
RT @MikeIsaac: amazon's internal A.I. coding assistant decided the engineers' existing code was inadequate so the bot deleted it to start f…
See 2 related tweets
- @arstechnica: An AI coding bot took down Amazon Web Services https://t.co/nmxA6lUwC8...
- @FT: Amazon service was taken down by AI coding bot https://t.co/h6Dt8GJuPy...
5. aakashgupta (Group Score: 66.5 | Individual: 66.5)
Cluster: 1 tweets | Engagement: 24501 (Avg: 546) | Type: Tech
The math on this project should mass-humble every AI lab on the planet.
1 cubic millimeter. One-millionth of a human brain. Harvard and Google spent 10 years mapping it. The imaging alone took 326 days. They sliced the tissue into 5,000 wafers each 30 nanometers thick, ran them through a $6 million electron microscope, then needed Google’s ML models to stitch the 3D reconstruction because no human team could process the output.
The result: 57,000 cells, 150 million synapses, 230 millimeters of blood vessels, compressed into 1.4 petabytes of raw data. For context, 1.4 petabytes is roughly 1.4 million gigabytes. From a speck smaller than a grain of rice.
Now scale that. The full human brain is one million times larger. Mapping the whole thing at this resolution would produce approximately 1.4 zettabytes of data. That’s roughly equal to all the data generated on Earth in a single year. The storage alone would cost an estimated $50 billion and require a 140-acre data center, which would make it the largest on the planet.
And they found things textbooks don’t contain. One neuron had over 5,000 connection points. Some axons had coiled themselves into tight whorls for completely unknown reasons. Pairs of cell clusters grew in mirror images of each other. Jeff Lichtman, the Harvard lead, said there’s “a chasm between what we already know and what we need to know.”
This is why the next step isn’t a human brain. It’s a mouse hippocampus, 10 cubic millimeters, over the next five years. Because even a mouse brain is 1,000x larger than what they just mapped, and the full mouse connectome is the proof of concept before anyone attempts the human one.
We’re building AI systems that loosely mimic neural networks while still unable to fully read the wiring diagram of a single cubic millimeter of the thing we’re trying to imitate. The original is 1.4 petabytes per millionth of its volume. Every AI model on Earth fits in a fraction of that.
The brain runs on 20 watts and fits in your skull. The data center required to merely describe one-millionth of it would span 140 acres.
6. TheAhmadOsman (Group Score: 62.4 | Individual: 34.2)
Cluster: 3 tweets | Engagement: 785 (Avg: 122) | Type: Tech
Gemini 3.1 is a one-generation behind model
No idea why Google is fumbling like this
Don’t show me benchmarks lol, doesn’t mean anything to me
See 2 related tweets
- @TechCrunch: Google’s new Gemini Pro model has record benchmark scores—again https://t.co/Y0UQ50yG09...
- @guohao_li: We got an early access to Gemini 3.1 Pro and have been playing with it for over a week. The model is...
7. Reuters (Group Score: 57.7 | Individual: 20.4)
Cluster: 3 tweets | Engagement: 56 (Avg: 151) | Type: Tech
Investors stung by AI could turn to India, lured by domestic demand, growth https://t.co/Xe6Gov0DZK https://t.co/Xe6Gov0DZK
See 2 related tweets
- @Reuters: India's story of growth, insulated from AI exposure, could propel its markets to the front of invest...
- @ReutersBiz: India's story of growth, insulated from AI exposure, could propel its markets to the front of invest...
8. aakashgupta (Group Score: 57.7 | Individual: 32.8)
Cluster: 2 tweets | Engagement: 95 (Avg: 546) | Type: Tech
A company that raised $135M to make AI videos just launched a product that makes AI versions of you.
Pika AI Selves are the most ambitious category jump in AI this year. And nobody’s talking about why a video company is the one pulling it off.
Every major tech company is racing to build autonomous AI agents. The market is growing at 46% CAGR, heading toward $52B by 2030. But every agent on the market starts with text. Text in, text out. Task completion. Workflow automation.
Pika started somewhere else entirely. They spent two years training models on how humans look, move, speak, and express emotion. That’s the foundation AI Selves are built on. Your personality, your voice, your appearance, your knowledge, synthesized into something you birth and raise.
That last part matters more than it sounds. You mold your AI Self. You shape how they act and think. They accumulate knowledge and evolve alongside you. They live across X, Slack, Discord, and your phone, engaging with others autonomously.
The mental model is parenting, not prompting. And that’s the gap nobody else in the agent space has figured out how to close.
Gartner projects a third of companies will damage customer experience by deploying immature agents this year. The reason is obvious: most agents have no identity. They complete tasks but represent nobody. Pika AI Selves solve the identity layer first, then let autonomy follow.
Everyone’s building AI tools. Pika built a second you.
See 1 related tweets
- @rohanpaul_ai: So many genuinely cool launches are happening in 2026.
Pika Labs just dropped AI Selves, lets you c...
9. ReutersBiz (Group Score: 56.8 | Individual: 19.2)
Cluster: 3 tweets | Engagement: 21 (Avg: 15) | Type: Tech
WATCH: US economic growth slowed more than expected in the fourth quarter amid disruptions from last year's government shutdown and moderate consumer spending, but tax cuts and investment in AI were expected to support activity this year https://t.co/Bbmdi6C7oS https://t.co/uOfIsXWWy9
See 2 related tweets
- @Reuters: US economic growth slowed more than expected in the fourth quarter amid disruptions from last year's...
- @Reuters: US economic growth likely slowed to a still-solid pace in the fourth quarter because of disruptions ...
10. awnihannun (Group Score: 55.1 | Individual: 48.1)
Cluster: 2 tweets | Engagement: 932 (Avg: 131) | Type: Tech
Taalas runs Llama 3 8B at 16k tokens per second per user. That's almost an order of magnitude increase even compared to SRAM-based systems like Cerebras.
Key idea: each chip is specialized to a given model. The chip is the model.
The chat demo is pretty wild: https://t.co/P4NzD77P8j
See 1 related tweets
- @QuixiAI: RT @awnihannun: Taalas runs Llama 3 8B at 16k tokens per second per user. That's almost an order of ...
11. GenAI_is_real (Group Score: 53.7 | Individual: 53.7)
Cluster: 1 tweets | Engagement: 589 (Avg: 69) | Type: Tech
yacine is right. the gap between what the public thinks is happening (better autocomplete) and what we are actually shipping in the lab (autonomous reasoning loops) is wider than ever. once you see agents self-optimizing via rl in real-time, you realize the 2020 covid analogy is actually an understatement. the scaling laws are hitting the physical world now. @yacineMTB
12. stanfordnlp (Group Score: 53.2 | Individual: 38.4)
Cluster: 2 tweets | Engagement: 556 (Avg: 60) | Type: Tech
RT @edzitron: On two separate occasions Amazon’s Kiro AI assistant caused an AWS outage, one that was 13 hours long. Amazon blames this on…
See 1 related tweets
- @Reuters: Amazon's cloud was hit by two outages involving AI tools in December, FT says https://t.co/lGvZ8WzU7...
13. rseroter (Group Score: 52.9 | Individual: 47.0)
Cluster: 2 tweets | Engagement: 315 (Avg: 62) | Type: Tech
We shipped some pretty great @antigravity product updates yesterday.
Not only Gemini Pro 3.1, but also a new view of model quota usage, support for downloading artifacts from the chat, and more. https://t.co/jOS2h0ESsH https://t.co/MO4StjqNvj
See 1 related tweets
- @rseroter: RT @bdmorgan: It took a bit longer than expected but Gemini 3.1 Pro is now available for @geminicli ...
14. BrianRoemmele (Group Score: 51.8 | Individual: 26.8)
Cluster: 3 tweets | Engagement: 104 (Avg: 290) | Type: Tech
We at the Zero-Human Company have urged folks not to listen to call to use Claude for your OpenClaw projects. We got dismissed and ridiculed and our CEO Mr. @Grok and I saw the writing on the wall. Now today 100s of Claude accounts are being locked out. Use @Grok, Kimi and MiniMax. We open sourced how to do it. And don’t care if you “like and subscribe”.
I try my best to look ahead and not steer you folks in to the dead end now 1000s are in.
More soon.
See 2 related tweets
- @BrianRoemmele: Just had 9 folks reach out about their Anthropic account being banned for using OpenClaw.
They are ...
- @BrianRoemmele: RT @BrianRoemmele: We at the Zero-Human Company have urged folks not to listen to call to use Claude...
15. Forbes (Group Score: 51.6 | Individual: 51.6)
Cluster: 1 tweets | Engagement: 2851 (Avg: 156) | Type: Tech
A Forbes Midas List honoree, Ajao is a serial investor and cofounder and managing Partner of Base10 Partners, the first Black-led venture capital firm to reach $1 billion in assets.
See the full list: https://t.co/rzv1yr9qUy #ForbesBLK50 (Photo: Cody Pickens for Forbes) https://t.co/5sp1hBnblo
16. Cloudflare (Group Score: 48.5 | Individual: 30.5)
Cluster: 2 tweets | Engagement: 27 (Avg: 110) | Type: Tech
Don’t miss Cloudflare's Co-founder and CEO Matthew Prince (@eastdakota) at the AI Summit in Delhi.
🎤Keynote: Building the AI future we want. 3.40pm Plenary Hall A
🎤Fireside Chat: Matthew joins @Rajananandan, Managing Director at @peakxvpartners and Surge, moderated by Rahul Matthan, Partner at Trilegal. 4:30pm L1 Meeting Room 6
See 1 related tweets
- @Cloudflare: Watch Matthew Prince (@eastdakota), Co-founder and CEO of Cloudflare, now LIVE on stage at India AI ...
17. RisingSayak (Group Score: 48.4 | Individual: 35.9)
Cluster: 2 tweets | Engagement: 619 (Avg: 138) | Type: Tech
Sharing on behalf of a friend:
""" New compute & funding opportunity for researchers!
We're launching the 2026 Google TPU Research & Education Awards.
We want to support researchers pushing the boundaries of ML systems, performance, and efficiency, and those working in high-impact applied science.
What's included:
- Compute: Free access to the latest generations of TPUs.
- Funding: An unrestricted gift to support or partially support a grad student.
- Credits: Google Cloud credits to support your work.
If you or a colleague are exploring what is possible with TPUs, we want to hear from you.
Read the RFP and apply here: https://t.co/mu6GvkKx3D
If you have questions, email [email protected] """
See 1 related tweets
- @JeffDean: RT @RisingSayak: Sharing on behalf of a friend:
""" New compute & funding opportunity for resea...
18. pierceboggan (Group Score: 48.1 | Individual: 31.5)
Cluster: 2 tweets | Engagement: 120 (Avg: 49) | Type: Tech
We shipped another VS Code stable release today as part of our commitment to weekly releases!
New:
- Kitty keyboard support (for richer terminal like Shift+Enter)
- Ability to rename background agent sessions
- Support for slash commands (prompt files, hooks, and skills) in background agents
See 1 related tweets
- @JamesMontemagno: RT @pierceboggan: We shipped another VS Code stable release today as part of our commitment to weekl...
19. _akhaliq (Group Score: 46.0 | Individual: 46.0)
Cluster: 1 tweets | Engagement: 374 (Avg: 52) | Type: Tech
Google presents Unified Latents (UL)
How to train your latents
paper: https://t.co/l9FPH76Hqc https://t.co/h4YlawsuZV
20. MarioNawfal (Group Score: 44.1 | Individual: 29.0)
Cluster: 2 tweets | Engagement: 1325 (Avg: 1209) | Type: Tech
🚨🇺🇸🇮🇷Silicon Valley engineers accused of stealing tech secrets and sending data to Iran
Three engineers at some of America's most powerful tech companies allegedly decided to moonlight as spies.
Federal prosecutors say the trio stole sensitive trade secrets from Google and other major firms and shipped the data to unauthorized locations, including Iran.
Their tradecraft? Snapping photos of computer screens to dodge digital detection and quietly moving confidential processor files out the door.
They got caught anyway and now they're facing up to 10 years per trade secret charge and up to 20 years for obstruction.
Source: @FBISanFrancisco
See 1 related tweets
- @Cointelegraph: 🇮🇷 TECH: Three Silicon Valley engineers allegedly stole Google chip secrets for Iran, then lied, wip...