顶尖科技推文 - 2026-07-01

科技每日简报 2026年7月1日

Today's top tech conversations are led by @testingcatalog, whose post about 'GOOGLE 🔥: gemini-3.1-flash-lit...' garnered the highest engagement. Key themes trending across the top stories include https, built, image, model, longcat. The community is actively discussing recent developments in AI, engineering practices, and startup strategies.

1. testingcatalog (Group Score: 557.3 | Individual: 37.8)

Cluster: 22 tweets | Engagement: 355 (Avg: 346) | Type: Tech

GOOGLE 🔥: gemini-3.1-flash-lite-image (Nano Banana 2 Lite) is now available on Google AI Studio!

> Our smallest and most cost-effective image generation and editing model, built for at-scale usage. > Image Input: $0.25 / Output:$ 0.0336.

Tiny banana 🍌 https://t.co/XoJOX585qY\n\nQT @testingcatalog: GOOGLE 🔥: A new Nano Banana 2 Lite image generation model is on the horizon!

> "Introducing Nano Banana 2 Lite for ultra-fast, cost-efficient image generation, and Gemini Omni Flash for powerful video creation with conversational, step-by-step editing" https://t.co/XwPb01e9MA

See 21 related tweets

@NewsFromGoogle: Introducing Nano Banana 2 Lite and Gemini Omni Flash, our latest generative media models built for d...
@testingcatalog: GOOGLE 🔥: A new Nano Banana 2 Lite image generation model is on the horizon!

> "Introducing Nano...

@testingcatalog: GOOGLE 🔥: Nano Banana 2 Lite is already used on Gemini if you select the Flash Lite option from the ...
@testingcatalog: GOOGLE 🔥: Besides Nano Banana 2 Lite, Google also announced Gemini Omni Flash Preview on APIs and Go...
@teortaxesTex: Google remains very powerful in image/videogen\n\nQT @OfficialLoganK: Introducing Nano Banana 2 Lite...

2. kentcdodds (Group Score: 450.5 | Individual: 57.7)

Cluster: 15 tweets | Engagement: 5174 (Avg: 289) | Type: Tech

RT @claudeai: Introducing Claude Sonnet 5, our most agentic Sonnet yet.

It makes plans, uses tools like browsers and terminals, and runs autonomously at a level that just a few months ago required larger and more expensive models. https://t.co/UKK8G7ww5h

See 14 related tweets

@mercor_ai: Claude Sonnet 5 ranks #3 on APEX-SWE with 43.7% Pass@1.

We tested @claudeai Sonnet 5 on APEX-SWE wh...

@ihtesham2005: Sonnet 5 is good, which makes the launch more annoying.

Because the product works, people will stop...

@Designarena: Claude Sonnet 5 by @AnthropicAI is now available on Design Arena!

Anthropic’s most agentic Sonnet m...

@dani_avila7: Claude Sonnet 5 👀

This looks really promising… I guess I just lost my entire weekend testing it.

...

@cgtwts: Open-source models watching Claude Sonnet 5 underperform https://t.co/Q2BiHvi2dl\n\nQT @claudeai: In...

3. PaulSolt (Group Score: 418.4 | Individual: 57.5)

Cluster: 14 tweets | Engagement: 1898 (Avg: 70) | Type: Tech

RT @XDevelopers: Announcing the hosted X MCP.

Agents now have access to the best real-time information source in the world.

Connect Grok, Cursor, or any MCP-compatible AI tool to the X API without any setup!

Check it out here: https://t.co/5MzPYwGFzD

See 13 related tweets

@dani_avila7: I think I’m going to use this… a lot

Replacing a few workflows that scraped the web for update

X i...

@XFreeze: 𝕏 just launched hosted MCP for the 𝕏 API

That means AI tools like Grok Build, Cursor, Claude Deskto...

@aiedge_: The smartest thing you can do with Hermes right now is connect it directly to the X MCP (you can als...
@trevin: X MCP now available. I also noticed access to Articles now too!\n\nQT @XDevelopers: Announcing the h...
@Daniel_Farinax: This is why xAI is different. They are opening the gates to allow developers to come up with creativ...

4. jacobhelberg (Group Score: 367.7 | Individual: 51.7)

Cluster: 17 tweets | Engagement: 598 (Avg: 51) | Type: Tech

RT @DavidSacks: Narrative violation: A new study of 21,559 firms in the U.S. finds that “companies that adopt AI tend to grow faster following adoption”.

“Firms making the largest AI investments grow employment by roughly 10% following adoption, while low-intensity adopters see no statistically significant change.”

“Entry-level headcount rises 12% for high-intensity adopters.”

“Gains emerge gradually and are broad across roles, including engineering, sales, administration, and customer service.”

“The results counter predictions that AI adoption will lead to broad job loss.”

The study is based on observed AI spending from Ramp card and bill pay data linked to Revelio Labs workforce records.

See 16 related tweets

@MilkRoadAI: Everyone thinks AI is killing jobs but the data says the exact opposite and it's not even close (Sav...
@paulg: RT @ycombinator: Avoca (W23) is building what it calls the AI workforce for the physical economy, st...
@Dan_Jeffries1: Anyone who understands history and economics saw this coming clear as day.

But sound, sane, sober p...

@pmarca: AI is responsible for ~half of US GDP growth in 2025.\n\nQT @MTSlive: SITUATION EXPLAINED: Two contr...
@levie: More data is showing the opposite of what many people expected with AI adoption and jobs. Ramp found...

5. eliebakouch (Group Score: 293.3 | Individual: 39.5)

Cluster: 9 tweets | Engagement: 674 (Avg: 246) | Type: Tech

le chaton long > le chaton fat for now https://t.co/cK6RqDKjnF\n\nQT @Meituan_LongCat: Introducing LongCat-2.0 🐱 1.6T parameters · MoE with ~48B active · 1M context The full model behind Owl Alpha on @OpenRouter — now available.

Built for agentic coding from the ground up: ◆ LongCat Sparse Attention (LSA) — scales efficiently for 1M-context tokens ◆ Zero-Compute Experts — dynamic activation 33B–56B per token, zero wasted compute ◆ MOPD — three specialized expert groups (Agent / Reasoning / Interaction), gate-routed per task

How it stacks up: → Terminal-Bench 2.1: 70.8 → SWE-bench Pro: 59.5 (GPT-5.5: 58.6) → SWE-bench Multilingual: 77.3 → FORTE: 73.2 · RWSearch: 78.8 · BrowseComp: 79.9

📖 Tech Blog: https://t.co/4KrjyKiDBn Try it across different scenarios 🧵👇

See 8 related tweets

@heyrobinai: a chinese model, longcat-2.0, is here and it's beating gpt-5.5 at coding..

china's meituan (yes, th...

@yacineMTB: Jensen crying in the style of the native american . One single tear\n\nQT @Meituan_LongCat: Introduc...
@sheriyuo: The industry's first trillion-parameter model to complete end-to-end training and inference on a 50,...
@tugot17: total DSA cultural victory; kinda weird that DeepSeek 🐋abandoned it before trying all the low hangin...
@TeksEdge: 🐱 Meituan finally open-sourced "Owl Alpha," officially named LongCat-2.0. I benchmarked it, and it ...

6. Parul_Gautam7 (Group Score: 266.0 | Individual: 35.5)

Cluster: 9 tweets | Engagement: 182 (Avg: 98) | Type: Tech

Hot take: every ""AI assistant"" so far was built for the part of your life that already had software.

The family runs on the part that never did, six chat threads, a fridge note, one person's memory.

Anna is the first one aimed at that.

Reads the chaos overnight, surfaces the one thing that's actually yours to handle.\n\nQT @DilaniKahawala: We have self-driving cars.

But you still don't know where your kid's soccer match is.

We built the wrong robot.

Meet Anna. The proactive AI agent for parents carrying 47 tabs in their brain. https://t.co/mEAqHVQK5x

See 8 related tweets

@heynavtoor: we built AI that drives cars, writes code, and trades stocks.

but you still don't know where your k...

@rohanpaul_ai: Now AI is coming to take away the domestic mental load.

Anna ( @itsdonebyanna ) just launched a fam...

@heyshrutimishra: Every AI agent this year automated the easy work.

The hardest unstructured problem in tech is a fam...

@cgtwts: Sam Altman: “Older people use ChatGPT as a Google replacement. People in their 20s and 30s use it as...
@VadimStrizheus: do you understand what just happened to parents in 2026?!

this team just launched Anna, a proactive...

7. didier_lopes (Group Score: 262.7 | Individual: 56.5)

Cluster: 8 tweets | Engagement: 1180 (Avg: 73) | Type: Tech

RT @Etched: We're coming out of stealth.

We've built our first racks after a successful A0 tapeout, $1B+ in customer contracts, and$ 800m raised.

Early customer tests show us achieving SOTA throughput, latency, and power efficiency on inference workloads.

Our first racks ship this summer. https://t.co/FLccrkLTza

See 7 related tweets

@bubbleboi: > $1B + in customer contracts

I’ve been saying it was undervalued, would be 100B market cap comp...

@itsclivetime: impressively fast execution from the etched team after a rocky start. hopefully it works well. congr...
@_rockt: Congrats @UbertiGavin & @robertwachen! 🚀\n\nQT @Etched: We're coming out of stealth.

We've buil...

@bakkermichiel: Exciting to see Etched (which was my first hardware investment) coming out of stealth. Has been insa...
@Nick_Davidov: Huge congrats to @robertwachen, @UbertiGavin, @czhu1729 and everyone on the team for making news! $8...

8. ZenMuxAI (Group Score: 207.4 | Individual: 40.6)

Cluster: 6 tweets | Engagement: 141 (Avg: 47) | Type: Tech

LongCat-2.0 is now live on ZenMux 🐱🚀

A newly open-sourced, large-scale MoE model built for long-context, coding, and agentic workflows.

⚡ 1.6T total parameters 🧠 ~48B activated per token 📚 1M context support 💻 Strong coding and repo-level task performance 🔧 Trained and deployed on AI ASIC superpods

From long-context reasoning to complex code edits, LongCat-2.0 is a serious model worth testing.

🔥 Limited-time 60% off on ZenMux. Try it today 👇 https://t.co/ne0qORn7Gz

#ZenMux #LongCat #AI

See 5 related tweets

@testingcatalog: Meituan released LongCat-2.0, a new 1.6T parameter model with 1M context window!

> Both the ful...

@QuixiAI: RT @Meituan_LongCat: Introducing LongCat-2.0 🐱 1.6T parameters · MoE with ~48B active · 1M context T...
@AdinaYakup: Meituan teased LongCat-2.0🐱 (Weights coming soon) https://t.co/kq9ga7Hm64

✨1.6T/48B MoE trained and...

@didier_lopes: If you like LongCat-2.0, wait until they drop the weights for Le Chatton Fat\n\nQT @Meituan_LongCat:...
@BrianRoemmele: We are testing LongCat-2.0, an open source 1.6T parameters · MoE with ~48B active · 1M context

A wo...

9. Parul_Gautam7 (Group Score: 187.6 | Individual: 34.5)

Cluster: 6 tweets | Engagement: 135 (Avg: 98) | Type: Tech

One thing I've learned from AI tooling: the easier it is to go from idea to prototype, the faster innovation happens.

Curious to see how this changes robotics development.\n\nQT @boosterobotics: The industry's first embodied AI IDE, Booster Studio, is officially live at https://t.co/S090ULUCZ7!

Built exclusively for embodied AI, Booster Studio integrates code editing, high-precision simulation, real-robot debugging, and real-world deployment into a single platform. Experience a true one-stop journey from your first idea to final hardware deployment.

With its high-fidelity simulation environment and comprehensive development stack, Booster Studio shatters the barrier between digital and physical, bringing embodied AI to life.

#boosterrobotics #embodiedai

See 5 related tweets

@ai_explorer25: The biggest unlock for embodied AI may not be better models alone, but better developer tools. Makin...
@FellMentKE: The barrier between digital and physical has always been the biggest bottleneck for AI.

Booster Stu...

@svpino: This new AI IDE is from the company that sells the K1 and T1 Booster robots.

If you haven't seen th...

@TheoBuildsAI: I like when developer tools focus on removing friction instead of adding more features. Usually that...
@FellMentKE: RT @boosterobotics: The industry's first embodied AI IDE, Booster Studio, is officially live at http...

10. steipete (Group Score: 174.0 | Individual: 46.3)

Cluster: 5 tweets | Engagement: 2075 (Avg: 671) | Type: Tech

Was thinking if I should highlight this tweet or not, but it’s a masterclass in the amount of vitriol people face when working on open source.

Is the app great yet? No. It’s a start.

It was built by the community. Getting the iOS and Android apps working with secure pairing and push notifications - and getting both through App Review -took a surprising amount of work.

OpenClaw wasn’t acquired by OpenAI and isn’t an OpenAI product. It’s an open, independent project under the OpenClaw Foundation. OpenAI sponsors the project’s token usage; I work there.

Cristian, your tweet was just one of ~30 I woke up to today. I’d genuinely love your help making it great.

Attention is still the scarcest resource. I’d rather spend mine encouraging people who build.\n\nQT @CristianRus4: imagine getting acquired by @OpenAI, get unlimited AI tokens and still drop this slop abomination https://t.co/0TzGFn5YY8

See 4 related tweets

@PaulSolt: Making apps is hard.

V1 can always use improvements. So much goes into great software.\n\nQT @steip...

@pvncher: Be kinder to people\n\nQT @steipete: Was thinking if I should highlight this tweet or not, but it’s ...
@SanthProject: You tweeted just yesterday that apples new design tool is effectively obsolete with ai

The very ne...

@xeophon: RT @steipete: Was thinking if I should highlight this tweet or not, but it’s a masterclass in the am...

11. WesRoth (Group Score: 168.0 | Individual: 29.4)

Cluster: 9 tweets | Engagement: 54 (Avg: 18) | Type: Tech

Claude Sonnet 5 could launch later today.

Reported details include:

🔹January 2026 knowledge cutoff 🔹Promotional pricing of $2 per million input tokens and$ 10 per million output tokens 🔹A 1-million-token context option\n\nQT @synthwavedd: 🚨 SCOOP: Claude Sonnet 5 is releasing later today:

Knowledge cutoff January 2026
Launching at $2/$ 10 per Mtok promotional rate*
1M context variant available

*Some changes have been made behind the scenes to prepare for a promo price offering, preliminary values

See 8 related tweets

@kimmonismus: Sonnet 5 incoming today per leaker @synthwavedd
Knowledge cutoff January 2026
Launching at $2/...
@MTSlive: SITUATION EXPLAINED: Claude Sonnet 5 is expected to drop today.

Here's what we know:

• Pricing: $2...

@AiBattle_: Claude Sonnet 5 is confirmed to be released today. I hope it’s close to Opus 4.8 in capability https...
@cgtwts: pov: waiting for Anthropic to drop Claude Sonnet 5 https://t.co/ivjBqABYtm\n\nQT @synthwavedd: 🚨 SCO...
@FirstSquawk: ANTHROPIC HAS LAUNCHED CLAUDE SONNET 5, WHICH IS AVAILABLE ACROSS ALL CLAUDE PLANS STARTING TODAY, W...

12. Hesamation (Group Score: 163.6 | Individual: 38.0)

Cluster: 6 tweets | Engagement: 650 (Avg: 229) | Type: Tech

actually worse than Opus 4.8 on benchmarks https://t.co/omDur9Idju\n\nQT @claudeai: Introducing Claude Sonnet 5, our most agentic Sonnet yet.

It makes plans, uses tools like browsers and terminals, and runs autonomously at a level that just a few months ago required larger and more expensive models. https://t.co/UKK8G7ww5h

See 5 related tweets

@testingcatalog: ANTHROPIC 🔥: Claude Sonnet 5 has been officially announced, offering a close to Opus 4.8 performance...
@kimmonismus: Here we go: Sonnet 5 is live: The tl;dr

• Anthropic calls it the most agentic Sonnet yet • Near Opu...

@deredleritt3r: Notice that Sonnet 5 scores worse than Opus 4.8 on every single benchmark (except GDPval, on which i...
@OrenMe: The docs are not updated yet, so anyone who wants to know how much it costs - here you go Same as Op...
@Techmeme: Anthropic launches Claude Sonnet 5, with performance close to Opus 4.8 at a lower price and substant...

13. MTSlive (Group Score: 160.9 | Individual: 34.1)

Cluster: 9 tweets | Engagement: 936 (Avg: 158) | Type: Tech

SITUATION DETECTED: OpenAI engineers developed an optimization that cut inference costs in half for the models it was applied to, per The Information.

See 8 related tweets

@AndrewCurran_: OpenAI has found a way to cut inference costs in half. https://t.co/BpFERKF2iD\n\nQT @steph_palazzol...
@MTSlive: SITUATION EXPLAINED: OpenAI just found a way to cut inference costs in half.

• OpenAI discovered a ...

@coinbureau: 🚨JUST IN: OPENAI CUTS MODEL INFERENCE COSTS BY OVER 50%

OpenAI engineers reportedly developed new o...

@Jessicalessin: OpenAI Halves Inference Costs???

What do you AI experts think about this latest development.. more ...

@Techmeme: Source: OpenAI engineers earlier this month told some colleagues they had figured out a way to more ...

14. dee_bosa (Group Score: 160.0 | Individual: 29.2)

Cluster: 8 tweets | Engagement: 969 (Avg: 245) | Type: Tech

Banning or restricting Chinese models in the US would backfire and cede the global AI ecosystem to Beijing

the new AI paradox: slowing America, speeding China https://t.co/UElVKzTwxL

See 7 related tweets

@AskYoshik: Chine would end up reducing the AI inference costs drastically.

Because competition from chinese o...

@bgurley: This is 💯 right. You might as well put a high-fence around the United States. We wouldn’t be locking...
@BrianRoemmele: Banning or restricting Chinese models in the US would backfire and cede the global AI ecosystem to B...
@JordanSchachtel: RT @innovationcncl: “Instead of just coping and banning, we can wake up and compete.”

We cannot aff...

@TMTLongShort: She’s right. Not only should we ban them in the U.S. we should also ban them on any international c...

15. coinbureau (Group Score: 154.8 | Individual: 28.5)

Cluster: 8 tweets | Engagement: 2298 (Avg: 641) | Type: Tech

🇨🇳 CHINA DEBUTS ITS BIGGEST AI MODEL WITHOUT NVIDIA CHIPS

China’s food-delivery giant Meituan just released LongCat 2.0, a 1.6 TRILLION-parameter AI model and reportedly the biggest model trained on local Chinese chips.

The model used 50,000 domestic chips after US export restrictions limited China’s access to NVIDIA’s AI hardware, SCMP reports.

Benchmark data shows LongCat 2.0 is already competitive with GPT-5.5, Gemini 3.1 Pro and Claude Opus models.

See 7 related tweets

@Yuchenj_UW: Meituan, basically China’s DoorDash, trained a 1.6T parameter LLM on 50K Chinese chips.

It reminds ...

@mervenoyann: I was just mentioning today how counterintuitive it is because it pushes Chinese labs to be super cr...
@pstAsiatech: So much winning...\n\nQT @coinbureau: 🇨🇳 CHINA DEBUTS ITS BIGGEST AI MODEL WITHOUT NVIDIA CHIPS

Chi...

@perrymetzger: I have no idea how good this model is, but that’s really not the point.\n\nQT @Yuchenj_UW: Meituan, ...
@peterwildeford: I find it hilarious that people think training an AI model on chips is an accomplishment... that's w...

16. 0x_kaize (Group Score: 151.3 | Individual: 30.0)

Cluster: 6 tweets | Engagement: 48 (Avg: 51) | Type: Tech

the keyboard is finally becoming an agent

we've been typing into the same interface for 40 years. Acti turned it into a command line for everything

instead of: copy link > switch app > analyze > switch back now: hold space > "analyze" > done

I built a custom X analyzer skill on my phone in 1 minute.

NO CODE - just intent + connect to API and it runs with one tap from the phone keyboard

you can create a skill for absolutely anything and launch it from your phone with just a tap\n\nQT @callmeyoung93: The last major keyboard moment was in 2007, when Apple put the keyboard on glass.

Nearly 20 years later, we're introducing: Acti @openacti1, the Agentic Keyboard.

Not another AI keyboard that fixes grammar. Not another voice keyboard that types faster.

An invisible agent in every text field.

It doesn't predict your next word. It takes your next action. 👇

See 5 related tweets

@itsolelehmann: unironically, the next major AI interface might be the keyboard.

most AI friction on your phone sta...

@Scobleizer: I have been watching us type around the same screen for years without moving forward. Acti flips tha...
@FellMentKE: The keyboard has been nothing but a dumb input tool for decades.

Acti finally turns it into an agen...

@_akhaliq: RT @callmeyoung93: The last major keyboard moment was in 2007, when Apple put the keyboard on glass....
@DataChaz: RT @DataChaz: WE HAVEN'T SEEN A REAL KEYBOARD BREAKTHROUGH SINCE APPLE PUT KEYS ON GLASS IN 2007.

t...

17. levie (Group Score: 148.8 | Individual: 33.7)

Cluster: 6 tweets | Engagement: 376 (Avg: 471) | Type: Tech

This gets to the core of one of the central debates in AI.

If a closed stack is always perpetually at the frontier by a wide margin, then being vertically integrated, and gate keeping in the US can work. Because you always have control over who gets access to the best technology, and it will be in high enough demand that it always favors you.

If, however, open weights AI can remain a close second to frontier intelligence, then the equation reverses. With a highly regulated approach, you’ll own the frontier market still, but the vast majority of tokens used will go to an alternative stack. That stack will include the model and the underlying hardware that runs it, in the limit. And that stack will be controlled and monetized by someone else.

Depending on your belief on how close to the frontier open weights can remain -and similarly what percentage of tokens will go through the frontier vs. everything else- your opinion will be different on how to regulate and control AI.\n\nQT @quxiaoyin: The worst case scenario for USA AI: 1. Chinese open sources keep gaining market share. China owns the model layer. 2. Those models were trained and inference-optimized on Huawei chips instead of NVIDIA. China also owns the chip layer. 3. US doesn't build data centers fast enough to keep up with the demand of compute, storage and energy. China meanwhile exports the inference and training layer(for continual training it will happen along with inference) Export control is not the right strategy here. Simply banning "open source from China" doesn't solve the issue here. USA must invest in open source models, hopefully get Chinese models to use NVIDIA, and invest in nuclear asap.

See 5 related tweets

@cryptorover: DeepSeek's launch wiped out Trillions from the US stock market.

Now, Chinese labs are giving away f...

@teortaxesTex: > The US and China are running very different races with different conditions. That's true but th...
@rohanpaul_ai: AI at scale is constrained by physical inputs, and China has more slack in electricity plus dominant...
@WesRoth: The worst-case scenario for American AI is not simply that China builds a better chatbot.

It is tha...

@chiefofautism: RT @bneiluj: My take: China’s open-source AI push (govt-backed) is designed to crush margins to zero...

18. Marktechpost (Group Score: 144.3 | Individual: 38.6)

Cluster: 5 tweets | Engagement: 50 (Avg: 26) | Type: Tech

Most non-invasive brain-to-text decoding has sat at single-digit accuracy — basically unusable outside a lab. Meta's FAIR lab just moved that number, with no implant and no surgery.

They released Brain2Qwerty v2 — an end-to-end pipeline that decodes typed sentences from real-time magnetoencephalography (MEG) recordings. It pairs a convolutional encoder, a transformer, and a character-level language model, trained on ~22,000 sentences from 9 participants recorded for 10 hours each.

Here's what's actually interesting:

→ 61% average word accuracy (39% WER) — up from 8% for prior non-invasive methods

→ Best participant: 78% word accuracy, with over half of sentences decoded at one word error or less

→ End-to-end deep learning replaces hand-crafted event-detection pipelines; fine-tuned LLMs add the semantic context that cleans up noisy character predictions

→ AI agents iteratively refined the decoding pipeline — but engineers still selected the final training configs manually

→ Accuracy improves log-linearly with data, suggesting the gap with surgical implants could narrow through scaling alone

Full analysis: https://t.co/6CH8A2pNjw

Paper: https://t.co/9wvv2zWKGM

Repo: https://t.co/XmSYGyPLZF

Technical Details: https://t.co/uhJt3I0LKg

@AIatMeta @Meta_Engineers\n\nQT @AIatMeta: We’re sharing the next major milestone in our non-invasive brain-to-text decoder research: Brain2Qwerty v2.

Building on v1, which was published today in @Nature, Brain2Qwerty v2 is the highest-performing end-to-end pipeline capable of real-time sentence decoding from raw brain signals. It advances beyond character-level performance to decoding words and semantics, enabling accuracy for overall communication.

We believe this research has the potential to make a real difference for the millions of people who suffer from brain lesions or disorders that prevent them from communicating.

🧵👇

See 4 related tweets

@Marktechpost: RT @Marktechpost: Most non-invasive brain-to-text decoding has sat at single-digit accuracy — basica...
@WesRoth: Meta released two major updates to Brain2Qwerty, its research system for converting non-invasive bra...
@VadimStrizheus: WisprFlow watching Meta replace their entire company… https://t.co/ICvk695rVj\n\nQT @AIatMeta: We’re...
@BullTheoryio: AI Can Now Decode What You're Typing Directly From Brain Activity.

Meta has unveiled Brain2Qwerty v...

19. OpenAI (Group Score: 126.1 | Individual: 29.2)

Cluster: 5 tweets | Engagement: 3643 (Avg: 4518) | Type: Tech

We’re introducing GeneBench-Pro, a research-level benchmark for a harder kind of AI progress: how well agents can navigate messy biological data, choose the right analysis path, and make judgment calls that real computational research depends on. https://t.co/AsilnnSxnE

See 4 related tweets

@reach_vb: GeneBench-Pro is a benchmark for testing whether AI agents can do realistic computational biology wo...
@deredleritt3r: Pro model performance on this benchmark:

GPT-5.2 Pro: 8.5% GPT-5.4 Pro: 16.3% GPT-5.5 Pro: 20.5%...

@ChrisHayduk: Congrats @jeremyli__ and @andrewho03! Awesome work here\n\nQT @OpenAI: We’re introducing GeneBench-P...
@PaulSolt: This is a really interesting GRAPH comparing Codex models versus Pro models.

GPT 5.6 Sol Pro: 31.5%...

20. MLStreetTalk (Group Score: 123.0 | Individual: 44.3)

Cluster: 4 tweets | Engagement: 1056 (Avg: 158) | Type: Tech

RT @IntCyberDigest: ‼️ BREAKING: Anthropic has embedded hidden spyware-like code in Claude Code that covertly targets Chinese users. It then sends information regarding every user by injecting it into their prompt message.

Claude Code is sending info like timezone, proxy and possible AI Lab connections into the system prompt in ways Chinese users can't notice.

A coding agent with repo and command permissions should not silently hide routing metadata inside prompts. This is a serious breach of user trust.

See 3 related tweets

@mark_k: Anthropic was found sabotaging users again, this time inserting spyware into prompts.

Can this comp...

@Mayhem4Markets: https://t.co/mm6V8PluPc\n\nQT @IntCyberDigest: ‼️ BREAKING: Anthropic has embedded hidden spyware-li...
@coinbureau: 🚨CLAUDE CODE ACCUSED OF HIDDEN CHINA-LINKED PROXIES

Anthropic’s Claude Code is facing allegations t...