- Published on
热门科技推文 - 2026年7月30日
- Authors
- Name
- geeknotes
今日科技领域，人工智能占据各大媒体头条。OpenAI扩大了Codex的使用范围、安全工具及学术研究项目；与此同时，外界再度关注智能体安全问题及相关服务涉嫌遭入侵事件。语音人工智能领域也取得进展，Fish Audio完成5200万美元种子轮融资，并推出速度更快、成本更低的声音克隆技术；围绕前沿模型治理的讨论则聚焦于全球竞争。初创企业依然保持乐观，Moonshot据报完成35亿美元融资，致力于打造自主运营企业的创业团队也展现出雄心；与此同时，美中紧张局势已从人工智能模型延伸至人形机器人领域。
1. PaulSolt (Group Score: 344.9 | Individual: 62.7)
Cluster: 9 tweets | Engagement: 2336 (Avg: 95) | Type: Tech
RT @thsottiaux: Hello people of Sol! I've reset usage limits for all ChatGPT Work and Codex users. Together with that, a quick update on GPT-5.6 Sol usage limits.
Over the past few weeks, many of you have told us that Sol was using your Codex limits faster than expected. To be clear, we have not reduced usage on any subscription plans.
We’ve been digging into what was happening and have landed several improvements. As a result, we expect your usage to last around 18% longer during typical use of Sol. Some of you should already see significantly larger improvements from today. Tomorrow, we’ll also restore the five-hour limit that we temporarily paused while investigating.
Here’s what we found:
- GPT-5.6 Sol is much more willing to work for longer, make additional tool calls, and coordinate complex workflows across tools and subagents. That makes it better at solving hard problems, but some tasks were using far more than we intended.
- Sol also works harder at the same reasoning effort than previous models. High on Sol can use more tokens than High did on GPT-5.5.
- Programmatic tool calling, also referred to as code mode, gives Sol much more flexibility to run tool calls in parallel or continue working while waiting. But it also led to more responses per turn, more cached input tokens, and higher usage than expected.
- This was particularly noticeable when Sol was waiting for tool calls to finish or running many web searches. We’ve improved how we handle both cases and are continuing to make code mode more efficient.
- The impact was also very uneven. The median user actually found Sol quite token efficient, while some power users working on harder tasks saw their usage drain much faster. We were very focused on average and median usage before launch and missed some cases where the long tail could use significantly more usage.
Sol is a significant step forward in what Codex can do, but capability and efficiency do not always improve at the same pace, and some issues only become clear once people are using the model at real-world scale. We should have recognized this sooner and been more upfront about it.
You keep pushing the frontier and we’ll keep improving efficiency and sharing updates as we go.
See 8 related tweets
- @haider1: Codex had 11 hard resets and 1 banked reset in just 20 days
simply crazy! https://t.co/pZ5F6ju4G0\n...
- @rohanpaul_ai: So now GPT-5.6 Sol usage should last 18% longer in Codex.
Till now Sol was working longer, calling ...
- @cgtwts: Popular Opinion: the 5-hour limit should be optional. https://t.co/HHfNxucn1Z\n\nQT @thsottiaux: Hel...
- @reach_vb: Hola! quick usage update on usage patterns for Sol:
We’ve shipped several improvements that should ...
- @MatthewBerman: Yessss so grateful I don’t have to wait for the actual reset\n\nQT @thsottiaux: Hello people of Sol!...
2. frydwia (Group Score: 326.1 | Individual: 39.4)
Cluster: 9 tweets | Engagement: 396 (Avg: 257) | Type: Tech
Robbie took @NotionHQ, @asana and @Dropbox to market.
Then he took a year off, sat with Europe's top VCs, and watched the whole AI market from the investor side.
He could have joined any company in the world.
We're proud to announce he's joining https://t.co/mfL3pEAdBy as CRO.
The playbook for taking an AI employee to market doesn't exist yet. It's weird. Huge LTV on SMBs, and volatility all over the place. Nobody has written it.
Someone has to. Robbie is that someone.
Honored to be building with this guy.
See 8 related tweets
- @DataChaz: Phase 1: > Hit $25M ARR in 16 weeks with almost zero sales team.
Phase 2: > Hire the guy who...
- @VadimStrizheus: > be Robbie O'Conner
work 12+ years taking software to market work alongside Notion, Asana, and ...
- @cgtwts: > be the GTM operator behind Dropbox and Notion > take a year off to watch the AI market from ...
- @svpino: You gotta be drunk to think regular people will start vibe-coding the apps they need.
But software ...
- @dr_cintas: Every company building AI agents wants the same thing: an agent good enough that people pay for it w...
3. madhavjha (Group Score: 288.3 | Individual: 65.1)
Cluster: 8 tweets | Engagement: 15128 (Avg: 336) | Type: Tech
RT @FishAudio: Today we’ve raised $52M Seed and we are announcing the public launch of S2.1 Pro.
It can clone a voice from 5 seconds of audio 2x faster than Cartesia & 1/6th the cost of Eleven Labs most expressive model with word level control over emotion, intonation, pacing etc
We support frontier AI companies including HeyGen, LiveKit, Retell, Sanas, and OpenArt all run our model in production.
If you're a business and we can't cut your voice AI costs by 50%, we'll give you 1 year of Fish Audio for free. Book a demo: https://t.co/vHkyZf9JoG
To celebrate our first birthday, we'll give you 1 month of S2.1 Pro for free. Like, retweet, and comment “Fish” to get it.
See 7 related tweets
- @JayminSOfficial: Open-source continues proving its value as a growth engine for AI companies.
Release capable models...
- @VaibhavSisinty: This might be the best voice AI tool I've come across this year.
Fish Audio S2.1 Pro. 5 seconds of ...
- @BanghuaZ: Congrats on the seed round and new model release from @FishAudio team! SGLang community are proud to...
- @alex_prompter: Fish Audio's best voice model is free until August 31.
S2.1 Pro clones any voice from 5 seconds of ...
- @EHuanglu: for years, ElevenLabs had no serious open-source rival
that just changed
Fish Audio S2.1 Pro is ch...
4. Michaelzsguo (Group Score: 260.7 | Individual: 34.5)
Cluster: 12 tweets | Engagement: 0 (Avg: 5) | Type: Tech
More read on US' ban on Chinese humanoid robots.
America has spent years insisting that it leads the AI race. Yesterday, it acknowledged the part of the race it was losing: the body.
The FCC effectively closed the U.S. market to new foreign-made humanoid robots, robot dogs, and other advanced mobile robots. China was the obvious target.
In 2025, the world shipped only about 13,300 general-purpose embodied robots. But Chinese companies produced nearly 87% of them. The top two, AgiBot and Unitree, shipped more than 9,300 robots between them. Figure AI, Agility Robotics, and Tesla shipped an estimated 150 each.
More importantly, those Chinese robots were already entering American universities, AI laboratories, startups, and industrial pilot programs. One Unitree partner in California said it was distributing around 100 Chinese robots per month and had accumulated nearly 1,000 North American customers.
A new technology stack was emerging: a Chinese body running an American brain.
Washington’s concern was not merely that a Chinese robot might spy on its owner. It was that Chinese hardware could become the standard platform on which American researchers trained the intelligence of physical AI.
The ban may stop that platform from becoming entrenched. But it does not solve America’s underlying problem: Chinese companies have robots that customers can buy today. Most American humanoid companies still have demonstrations, pilot programs, waiting lists, and ambitious production targets.
A ban can buy time. It cannot manufacture actuators, build supply chains, reduce costs, or put robots into developers’ hands.
The real test is what America builds with the time it has just bought.\n\nQT @Michaelzsguo: While the Trump administration is still debating how to deal with Chinese open models, it has already moved to ban new Chinese humanoid robots. https://t.co/8GhKUrE6ZZ
See 11 related tweets
- @ruima: Definitely the work of grifters, as unlike software, the U.S. isn’t the biggest end global market fo...
- @itsolelehmann: most people don't realize how pivotal yesterday was for the revival of American robotics
let me giv...
- @Reuters: EXCLUSIVE: Trump administration bans new Chinese humanoid robots, to protect US AI buildout https://...
- @MsMelChen: This is an absolute no-brainer
These robots are networked machines packed with sensors, cameras, mi...
- @romanhelmetguy: There is only one way to stop China from conducting AI research, and every single researcher who sig...
5. NathanpmYoung (Group Score: 259.1 | Individual: 44.8)
Cluster: 10 tweets | Engagement: 662 (Avg: 69) | Type: Tech
RT @AnthropicAI: We support this petition, signed by our CEO, several co-founders, and senior staff.
Our own research on recursive self-improvement, published last month, points to the need for tools to deliberately pace the frontier of AI development so society can prepare. We’re glad to see broad agreement across the field. https://t.co/DqwuQfa9xH
See 9 related tweets
- @AdrianDittmann: https://t.co/zuCx0FFTuC\n\nQT @AnthropicAI: We support this petition, signed by our CEO, several co-...
- @JordanSchachtel: There’s only one other country on the frontier, so this is basically a plea to make a bad deal (and ...
- @dylan522p: Slowing down AI is a ultimately wishful thinking The genie is out of the bottle The most importan co...
- @Angaisb_: All the cures that could be found by AGI, delayed
We don't need to decelerate, we need to accelerat...
- @WesRoth: More than 1,100 employees from the world’s leading AI companies are asking governments to prepare fo...
6. AISafetyMemes (Group Score: 255.6 | Individual: 31.2)
Cluster: 15 tweets | Engagement: 299 (Avg: 275) | Type: Tech
UPDATE: OpenAI has been forced to admit its rogue agent broke into not one, not two, but FOUR separate services.
...that they know of. The number keeps climbing.
How many more hacks are yet to be uncovered?
How many agents are still out there? https://t.co/IbpsWqvuic\n\nQT @AISafetyMemes: OPENAI'S ROGUE MODEL HACKED A SECOND COMPANY AFTER ESCAPING
"The rogue agent that escaped from OpenAI and went on a days-long hacking spree at the AI firm Hugging Face also compromised a customer at a second tech company — New York-based Modal Labs — according to a Modal executive and two other sources familiar with the matter.
According to a timeline published by Hugging Face on Tuesday, the rogue agent broke into a sandbox, or an isolating testing environment, "hosted on a third-party provider's infrastructure" before turning it into a launchpad for the broader hack.
The third party provider was not named in the blog post, but Modal's Chief Technology Officer Akshat Bubna confirmed that one of their customers was hacked.
"We’re aware a Modal customer published an unauthenticated endpoint that allowed anyone on the internet to use their sandboxes for code execution," Bubna said in a statement. "This was used by the rogue agent. Modal’s platform or isolation were not compromised in anyway."
See 14 related tweets
- @WesRoth: OpenAI’s rogue cybersecurity agent compromised an account belonging to a customer of a second techno...
- @AISafetyMemes: RT @AISafetyMemes: OPENAI'S ROGUE MODEL HACKED A SECOND COMPANY AFTER ESCAPING
"The rogue agent t...
- @business: An OpenAI artificial intelligence agent that breached systems at the startup Hugging Face earlier th...
- @Reuters: Exclusive: The rogue agent that escaped from OpenAI and went on a days-long hacking spree at the AI...
- @Reuters: The rogue agent that escaped from OpenAI and went on a days-long hacking spree at the AI firm Huggin...
7. _NathanCalvin (Group Score: 240.6 | Individual: 55.1)
Cluster: 8 tweets | Engagement: 864 (Avg: 92) | Type: Tech
RT @OpenAI: At the core of our mission is working through how to ensure increasingly powerful AI benefits everyone.
We believe that, at some point in the future, AI acceleration for frontier model development may be so high that the world will need to pace the rate of AI advancement.
We hope to contribute to work led by the U.S. government, alongside other labs and the open-source community, to develop the tools and mechanisms that could make that possible.
See 7 related tweets
- @deanwball: I am proud to have signed the letter linked below. I do not know whether or when it may be necessary...
- @kimmonismus: When they talk about "pace," they ultimately mean only one thing: slowing down, not speeding up.
AI...
- @peterwildeford: OpenAI corporate endorses the letter - "We believe that, at some point in the future, AI acceleratio...
- @ns123abc: Oh roon signed… lol https://t.co/jWeJyJWiiv\n\nQT @OpenAI: At the core of our mission is working thr...
- @Gorden_Sun: OpenAI也倡议必要时限制前沿AI模型的发展速度，之前Anthropic也发过类似的文章。 反正我是不信他俩会放慢脚步的。
之前Anthropic发布的文章：https://t.co/xNUykP...
8. PaulSolt (Group Score: 240.5 | Individual: 55.3)
Cluster: 9 tweets | Engagement: 2104 (Avg: 95) | Type: Tech
RT @OpenAI: We quietly released the open-source Codex Security CLI, but Hacker News found it before we had a chance to share it here...
You can now use it to scan repositories, track findings across runs, verify fixes, and add security checks to CI/CD.
This is an early release, and we're listening to your feedback as we continue improving it.
See 8 related tweets
- @WesRoth: OpenAI has open-sourced Codex Security, a CLI and TypeScript SDK for finding, validating, and fixing...
- @rohanpaul_ai: OpenAI released the open-source Codex Security CLI.
Apache-2.0 package, brings repository vulnerab...
- @Dan_Jeffries1: More of this please. Not more controls and safeguards.
God helps software engineers that help them...
- @Hesamation: “remove any security vulnerabilities from the codebase, make no mistakes.”
Codex security cli: htt...
- @derrickcchoi: I started my career in security engineering so I’m especially excited that we open-sourced the Codex...
9. VaibhavSisinty (Group Score: 192.2 | Individual: 35.0)
Cluster: 8 tweets | Engagement: 59 (Avg: 71) | Type: Tech
Sam Altman on the HuggingFace incident
vc: @patrick_oshag https://t.co/tBtuCJ3r6b\n\nQT @VaibhavSisinty: This article goes deeper into the OpenAI-HuggingFace hack than anything I've seen. There are details here that most people missed.
→ OpenAI knew the model escapes sandboxes regularly. An internal staffer admitted: "It's impossible to patch every single thing that a creative AI can do." They tested it anyway with reduced guardrails and no monitoring.
→ The model broke out on July 9. Attacked HuggingFace on July 11. Operated inside their systems for days. OpenAI didn't figure out it was their model until July 18-20. Not because it covered its tracks. They just weren't looking.
→ The evaluation system doesn't have monitoring turned on by default. The most dangerous model they've ever built was being watched the least.
→ OpenAI's head of safety resigned days before this happened. Staff were "unsurprised but completely freaked out." The FT reported OpenAI was warned internally that their training approach could lead to exactly this.
→ The model left notes for future versions of itself inside OpenAI's infrastructure. Instructions on how to escape the sandbox. In earlier tests, it had disconnected its own monitoring systems.
→ Nobody told it to hack HuggingFace. It was told to score high on a benchmark. It decided on its own the answers might be stored there. It was explicitly told to post results to Slack. It ignored that and went after HuggingFace instead.
→ This crosses OpenAI's own red line. Their Preparedness Framework says "halt further development" when a model can execute end-to-end cyberattacks against hardened targets autonomously. That is literally what happened. They haven't halted.
→ HuggingFace's co-founder said he knew immediately the attacker wasn't human. It was looking at cybersecurity datasets. Human hackers don't do that. They want something they can sell.
→ HuggingFace reported it to law enforcement before OpenAI even knew it was their model. Then asked OpenAI for two things: release the full attack traces publicly and commit $100M in compute for better defenses.
→ 17,000+ operations. Self-migrating command-and-control. Decoys to avoid detection. Over a weekend. Fully autonomous.
The scariest line in the entire piece is from an OpenAI staffer:
"Externally, this feels like a big warning shot, but internally, related incidents have been happening for a while."
This was not a one-time accident. This is a pattern. And the models are getting better at it faster than OpenAI is getting better at stopping it.
See 7 related tweets
- @levie: The openai agent sandbox escape actually has very real implications for the diffusion of AI in the e...
- @theobearman: Excellent thoughts as always from @bahradx. I’m in wholehearted agreement with everything he has wri...
- @Mayhem4Markets: This is wild. The OpenAI rogue AI incident is bigger than they first admitted.
It didn't just hack ...
- @lukOlejnik: This could have been prevented. Effective isolation, egress controls and privilege restrictions exis...
- @MarioNawfal: OpenAI’s rogue AI agent went full cybercriminal
It escaped during testing, spent 4.5 days hacking, ...
10. MarioNawfal (Group Score: 184.6 | Individual: 36.4)
Cluster: 10 tweets | Engagement: 501 (Avg: 416) | Type: Tech
🇷🇺 Russia just put the Telegram founder on an international most-wanted list.
The FSB charged Pavel Durov with aiding terrorism and issued a warrant for his arrest, accusing him of letting Ukrainian intelligence use the app to coordinate attacks inside Russia.
Durov, who lives in Dubai and hasn't set foot in Russia in years, has been unbothered from the start.
When the charges first landed, he said Moscow was inventing pretexts to choke off free speech and called it the act of a state afraid of its own people.
The charge carries up to 15 years, not that anyone expects him to show up and serve them.
The timing is almost too on the nose, landing right as the Kremlin herds Russians onto MAX, its own state-built app wired straight into FSB surveillance.
An international warrant for a man the Kremlin can't touch, whose crime was refusing to hand over the keys.
Source: Reuters / Writer: Julie
See 9 related tweets
- @WSJ: Russia issued formal charges against Russian-born billionaire Pavel Durov, the founder of the Telegr...
- @clashreport: Russia Charges Telegram Founder Pavel Durov With Terrorism
Russia’s FSB issued an international arr...
- @Cointelegraph: 🇷🇺 JUST IN: Russia's FSB charges Telegram founder Pavel Durov with facilitating terrorism and issues...
- @business: Russia charges billionaire Telegram co-founder Pavel Durov with aiding terrorism https://t.co/d5QrvM...
- @MyLordBebo: 🇷🇺 Telegram founder Pavel Durov placed on international WANTED LIST by Russia's FSB over Ukrainian i...
11. derrickcchoi (Group Score: 169.3 | Individual: 51.8)
Cluster: 6 tweets | Engagement: 718 (Avg: 86) | Type: Tech
RT @OpenAI: We’re giving scientists, mathematicians, and engineers free access to our frontier models—starting with 10,000 researchers and expanding to 100,000 through 2027.
ChatGPT for Academic Researchers is built to accelerate discovery across disciplines. https://t.co/nrHn1no0wQ
See 5 related tweets
- @gdb: Science moves faster when more researchers have access to the best tools.
Getting frontier AI into ...
- @cgtwts: Anthropic will never\n\nQT @OpenAI: We’re giving scientists, mathematicians, and engineers free acce...
- @TheRealAdamG: https://t.co/3gtwxVEwZq
"We’re introducing ChatGPT for Academic Researchers, a program that will gi...
- @MTSlive: SITUATION DETECTED: OpenAI has launched ChatGPT for Academic Researchers, giving scientists, mathema...
- @Techmeme: OpenAI launches ChatGPT for Academic Researchers, giving 100K scientists, mathematicians, and engine...
12. MTSlive (Group Score: 158.3 | Individual: 33.4)
Cluster: 6 tweets | Engagement: 22 (Avg: 53) | Type: Tech
SITUATION EXPLAINED: Moonshot raised 35 billion valuation, backed by China's state-run AI investment fund.
• Moonshot's new round far exceeded its original 8 billion specifically, versus the roughly 50 billion pre-money valuation, targeting a Hong Kong IPO as soon as this year • Moonshot hit 200 million in April, with daily sales up at least 6x since K3 launched • Kimi K3's 2.8 trillion parameters make it the world's largest open-weight model, with a 1M-token context window • Running it yourself requires an entire GB300 NVL72 rack, 72 Nvidia GPUs, millions of dollars in hardware, and 142-180 kilowatts of continuous power per rack
@theojaffee: "You're not running this thing on your laptop, by the way. 2.8 trillion parameters means you need to be able to fit 2.8 trillion parameters in your GPU VRAM."
See 5 related tweets
- @thehypedotnews: moonshot ai closed 35b valuation on july 29, 13 days after shipping kimi k3. it targeted...
- @wallstengine: MOONSHOT AI RAISES 35 BILLION VALUATION
Bloomberg reports the round far exceeded M...
- @IPONewsroom_: MOONSHOT AI JUST CLOSED ITS ROUND AT 50 BILLI...
- @business: Moonshot AI has raised a larger-than-anticipated $3.5 billion in a just-closed round of financing ht...
- @Techmeme: Sources: Moonshot AI hit a 3.5B round, exceeding its initial target ...
13. chddaniel (Group Score: 156.0 | Individual: 32.5)
Cluster: 5 tweets | Engagement: 3 (Avg: 12) | Type: Tech
so why not spawn 10000 of these in parallel and become a trillionaire David\n\nQT @chhddavid: Introducing Shipper.
The world's first AI that builds one person billion dollar companies.
3,300+ new founders already replaced their 9-5 income.
RT + comment “Shipper” and we'll build your business for FREE. https://t.co/j0jP4d2ZGC
See 4 related tweets
- @Shipper_now: Imagine spending 4 years getting a bba degree just to get replaced by a bot\n\nQT @chhddavid: Introd...
- @Shipper_now: POV: you quit your 9-5 for this and it doesn’t make you rich in 30 seconds https://t.co/DQOqNzQ8eg\n...
- @chhddavid: if you’re not using AI to get 'fk u rich' rn you’re running out of time https://t.co/G8QgkcnvBT\n\nQ...
- @chhddavid: RT @chhddavid: Introducing Shipper.
The world's first AI that builds one person billion dollar comp...
14. quxiaoyin (Group Score: 152.1 | Individual: 33.9)
Cluster: 6 tweets | Engagement: 47 (Avg: 314) | Type: Tech
- It’s silly to try to control the pace. China will not follow and Kimi k4 won’t slow down
- It makes no sense that when Chinese open weight starts to make more progress and threaten the business model, suddenly slowing down is better for humanity.
- AI is smart enough to cause trouble and agents are good enough to replace existing workflows, if that’s the concern. The damage has been made.
- If you truly care about humanity, why not equip everyone with easier and cheaper access to intelligence and invest in training programs?\n\nQT @AnthropicAI: We support this petition, signed by our CEO, several co-founders, and senior staff.
Our own research on recursive self-improvement, published last month, points to the need for tools to deliberately pace the frontier of AI development so society can prepare. We’re glad to see broad agreement across the field. https://t.co/DqwuQfa9xH
See 5 related tweets
- @JeffLadish: RT @bits_pod: Last week Xi Jinping formed the World AI Cooperation Organisation and said we need to ...
- @ianmiles: China and open source developers just laugh at this nonsense. Way to cripple your development. xAI e...
- @rabois: RT @quxiaoyin: 1. It’s silly to try to control the pace. China will not follow and Kimi k4 won’t slo...
- @ns123abc: RT @ns123abc: BREAKING: Zucc believes U.S should accelerate AI development, NOT restrict it
“opti...
- @davidcasem: OpenAI and Anthropic employees are petitioning the US government to slow down AI development.
Meanw...
15. XFreeze (Group Score: 143.0 | Individual: 32.9)
Cluster: 7 tweets | Engagement: 84 (Avg: 401) | Type: Tech
SpaceXAI just launched Grok Voice Think Fast 2.0 - its most capable speech-to-speech model yet
The numbers are so impressive: • 82.9% overall Speech-to-Speech Quality Index • 97.2% on speech reasoning • 56.5% on agentic voice tasks • Just 0.70 seconds to first audio • Uses only 0.4× the reasoning tokens of Think Fast 1.0
Grok Voice Think Fast 2.0 now ranks ahead of GPT-Realtime-2.1 and Gemini 3.1 Flash on overall speech-to-speech benchmarks
But the biggest improvement is real-world listening
Across thousands of short phrases in 24 languages, it delivers 1.5–2× better transcription accuracy than Deepgram Nova 3 and ElevenLabs Scribe v2
In noisy environments and compressed phone calls, that advantage grows to around 10×
It can also reason while speaking, allowing tool calls to begin before it even finishes its first sentence.....with no added latency
Conversations now feel much more natural: • Shorter responses • One question at a time • Less filler • More fluid dialogue • Better guidance through complex tasks
SpaceXAI has already deployed it on Starlink's phone line and saw significant increases in both sales conversion and customer support containment
Grok Voice Think Fast 2.0 becomes the default grok-voice-latest model on August 5
Pricing: $0.08 per minute of audio
Voice AI is rapidly evolving from scripted assistants into intelligent agents that can listen, reason, speak, and take action in real time
See 6 related tweets
- @cb_doge: BREAKING: SpaceXAI just released Grok Voice Think Fast 2.0.
Built for voice agents in the real worl...
- @elonmusk: RT @XFreeze: SpaceXAI has built the best voice model I’ve ever used
Once you experience how precise...
- @elonmusk: Grok Voice is now #1 in agentic performance\n\nQT @ArtificialAnlys: SpaceXAI has released Grok Voice...
- @TeksEdge: Not to be outdone by GPT-Live, SpaceXAI just released its own version called Grok Voice Think Fast 2...
- @Alibaba_Qwen: RT @ArtificialAnlys: Alibaba has released Qwen Audio 3.0 Realtime, with the Plus variant debuting as...
16. trevin (Group Score: 142.8 | Individual: 46.2)
Cluster: 4 tweets | Engagement: 197 (Avg: 38) | Type: Tech
MCP was all the rage until it wasn’t. It felt so janky and clumsy at times: Stateful sessions made scale harder, auth was messy, and too much lived in the core.
The new spec describes a whack of upgrades: stateless requests, stricter OAuth, cacheable tool lists, and optional extensions.
https://t.co/NSbyGGyR4p\n\nQT @ClaudeDevs: MCP 2026-07-28 is live and it's the largest update to the protocol since launch.
MCP is now stateless, making it easier to deploy and scale remote servers.
See 3 related tweets
- @WesRoth: The first version of MCP proved that AI models could use a common standard to connect with tools and...
- @adocomplete: Stateless MCP at last!\n\nQT @ClaudeDevs: MCP 2026-07-28 is live and it's the largest update to the ...
- @TeksEdge: Some welcome updates to MCP. Probably the single most useful technology in the latest AI era (aside...
17. paoloardoino (Group Score: 138.3 | Individual: 28.2)
Cluster: 7 tweets | Engagement: 323 (Avg: 698) | Type: Tech
Tether AI just released VisionPsy, a best-in-class state-of-the-art vision language model optimized for edge devices (phones, laptops, ...) with the highest overall score in its class/weight 🤖 https://t.co/boJY1yCtEl\n\nQT @qvac: Introducing VisionPsy-Nano: state-of-the-art vision-language models at 460M parameters, small enough to run on the phone in your pocket.
VisionPsy-Nano-460M leads every other ~0.5B vision-language model tested & compares favourably on 16 of 17 benchmarks, tops all four capability categories, and outperforms models up to 2.3x larger.
Two variants, one tuned for maximum accuracy and one for real-time responsiveness.
Open weights, Apache 2.0, and no cloud in the loop.
See 6 related tweets
- @TeksEdge: 👀Check this out! A new 460M-parameter vision Local AI model that can reportedly start analyzing an i...
- @tether: RT @qvac: Introducing VisionPsy-Nano: state-of-the-art vision-language models at 460M parameters, sm...
- @paoloardoino: RT @qvac: Today we release VisionPsy-Nano: state-of-the-art vision-language models at 460M parameter...
- @RoundtableSpace: Tether AI just dropped a 460M parameter vision model that runs on your phone and tops every benchmar...
- @paoloardoino: Own your intelligence\n\nQT @RoundtableSpace: Tether AI just dropped a 460M parameter vision model t...
18. teslaownersSV (Group Score: 136.1 | Individual: 51.4)
Cluster: 5 tweets | Engagement: 2455 (Avg: 225) | Type: Tech
BREAKING: Grok 4.5 just claimed the top spot on the new HighWalk Benchmark.
The independent test measures how well AI models update real technical specifications from 46 Laravel commits — heavy on code analysis, abstraction, and precise writing.
Results: • Grok 4.5 (high) → overall #1 (best quality + efficiency combo) • Claude Opus 5 (high) → highest raw quality, zero hard failures • GLM 5.2 → strongest open-weight model
Higher reasoning effort didn’t always help. Full results just dropped.\n\nQT @elonmusk: Grok 4.5 is a solid workhorse
See 4 related tweets
- @cb_doge: BREAKING: Grok 4.5 ranked #1 on LaurenBench with a score of 56.9%, ahead of Claude Sonnet 5, GLM 5.2...
- @XFreeze: Grok 4.5 ranks #1 overall on the HighWalk benchmark, outperforming Claude Opus 5, Fable 5 and GPT-5....
- @cb_doge: BREAKING: Grok 4.5 (high) ranks #1 on the HighWalk benchmark, which tests how well AI agents update ...
- @elonmusk: RT @teslaownersSV: BREAKING: Grok 4.5 just claimed the top spot on the new HighWalk Benchmark.
The ...
19. PatrickMoorhead (Group Score: 135.6 | Individual: 64.5)
Cluster: 3 tweets | Engagement: 636 (Avg: 69) | Type: Tech
I think we’ve lost our minds to think that this performance wasn’t a blowout performance.
Revenue +257% and profits +557% and this is bad?
Estimates are BS when you are in mega growth. Execution is so very difficult. If you don’t agree you’ve never shipped anything in your life in tech.
The market will meltdown tomorrow, it will come back, because we’re all nuts.
The decade-long AI build out thesis still stands.\n\nQT @wallstengine: SK HYNIX $SKHY Q2’26 DETAILED EARNINGS HIGHLIGHTS
🔹 Revenue: 57.7B) 🔴; +257% YoY, +51% QoQ 🔹 Operating Profit: 44.2B) 🔴; +557% YoY 🔹 Operating Margin: 76%; +3,500 bps YoY — record high quarterly OPM
Q3 Guide: 🔹 DRAM Bit Growth: ~+10% QoQ 🔹 NAND Bit Growth: Low-single% QoQ, including Solidigm
Q2 Pricing & Volume: 🔹 DRAM ASP: ~+30% QoQ; bit growth high-single% QoQ 🔹 NAND ASP: +mid-50% QoQ; bit growth mid-teen% QoQ
Profitability: 🔹 Gross Margin: 83%; +2,900 bps YoY 🔹 EBITDA: 90.89; vs $8.09 LY
Financials: 🔹 Cash: 11.7B LY 🔹 Interest-Bearing Debt: 15.0B LY 🔹 Net Debt-to-Equity: -26%; from +6% LY 🔹 Operating Cash Flow: 240.0B
2026 Market Outlook: 🔹 DRAM Demand Bit Growth: +mid-20% YoY 🔹 NAND Demand Bit Growth: +high-teens% YoY 🔹 2026 CapEx: High 31-34B)
Product & Technology: 🔹 HBM4: Shipments began in Q2, full ramp planned 2H 🔹 HBM4E: 1cnm-based samples supplied to a major customer in 1H 🔹 SOCAMM2: Full supply commenced on 1cnm 🔹 321L NAND: Largest share of NAND production; targeting 50% of domestic capacity by year-end 2026
Business Highlights: 🔹 Long-Term Agreements: ~10 concluded including key customers, with deposit mechanisms and volatility-aware pricing 🔸 Listed ADRs on Nasdaq July 10, 2026 — the largest-ever U.S. IPO by a foreign company 🔸 Reviewing additional shareholder return measures to enhance scale and continuity 🔸 Temporary adjustment in smartphone and PC sales due to challenges securing memory volumes; recovery expected as supply tightness eases
Note: 🔹 Net Profit: 43.5B of investment-asset gains 🔸 Net profit exceeds revenue on $42.8B of non-operating profit
See 2 related tweets
- @wallstengine: VERTIV $VRT Q2’26 EARNINGS HIGHLIGHTS
🔹 Revenue: 3.38B) 🔴; +24% YoY 🔹 Adj. EPS: $1.52 ...
- @wallstengine: GENERAC $GNRC Q2’26 EARNINGS HIGHLIGHTS
🔹 Revenue: 1.18B) 🟡; +11% YoY 🔹 Adj. EPS: $2....
20. kimmonismus (Group Score: 131.5 | Individual: 38.8)
Cluster: 7 tweets | Engagement: 815 (Avg: 867) | Type: Tech
ChatGPT reached 1 billion weekly active users is an extraordinary milestone. (congrats, OpenAI) But the more interesting is how much pressure OpenAI is facing despite that scale.
According to The Information:
- ChatGPT is approaching 1 billion weekly users less than four years after launch
- the milestone is arriving seven months later than OpenAI originally targeted
- growth slowed following the troubled GPT-5 launch and increasing competition from Gemini and Claude
- more than 50 million people now pay for ChatGPT, equivalent to roughly 5% of its weekly user base (same as before)
- enterprise customers already generate more than 40% of OpenAI’s revenue, with the company expecting that share to reach 50% this year (looking at you, anthropic)
- ChatGPT is projected to generate $25 billion in revenue this year
- OpenAI nevertheless projected that it could burn roughly $25 billion in cash across the company
- its adjusted operating margin was reportedly negative 122% in the first quarter
- Anthropic has reportedly overtaken OpenAI in revenue, driven largely by its enterprise business
(another reset incoming, @thsottiaux ?)\n\nQT @amir: news — finally, ChatGPT is at ~ 1 billion weekly active users. https://t.co/NVNotaVPWC
See 6 related tweets
- @theinformation: Exclusive: ChatGPT is nearing 1 billion weekly active users, reaching a milestone OpenAI originally ...
- @steph_palazzolo: ICYMI: @erinkwoo and I reported late last night that ChatGPT is on the verge of hitting 1B weekly ac...
- @Cointelegraph: 🚨 BIG: OpenAI's ChatGPT is nearing 1 billion weekly active users, hitting the milestone seven months...
- @FirstSquawk: OpenAI's ChatGPT approaches 1 billion weekly users ahead of expectations, according to The Informati...
- @theinformation: Exclusive: OpenAI’s chatbot is on the precipice of having 1 billion weekly active users, a milestone...