Thinking Machines Releases Its First AI Model, Acknowledging Inspiration from Chinese Rivals

Thinking Machines, the startup led by former OpenAI CTO Mira Murati, has released its first AI model. According to a report from the Financial Times, the model was developed drawing on techniques from Chinese competitors.

The article does not make clear the model's specific performance benchmarks or its release scope (open weights or API-only) at this time. That said, it is unusual for a US AI startup to openly state it "drew inspiration from Chinese rivals," and the remark is drawing attention as an acknowledgment of how rapidly open-weight models have evolved. On Hacker News, it picked up a relatively modest 4 points.

Alphabet Stock Falls as Markets React to Reports of a Gemini 3.5 Pro Delay

Alphabet shares fell following a CNBC report. According to the report, the release of Google's next-generation flagship model, Gemini 3.5 Pro, is running behind schedule.

Gemini 3.5 Pro was presumably positioned as Google's flagship model to counter Anthropic's Claude and OpenAI's GPT, but the article leaves the specific reasons for the delay and any new release date unclear. Coming on the heels of recent reports that "Alphabet's AI model Gemini has vanished from major LLM rankings," the news underscores once again just how unforgiving the competitive landscape has become.

Alibaba's Qwen Audio 3.0 TTS Plus Tops the Artificial Analysis Speech Arena

Alibaba's Qwen Audio 3.0 TTS Plus has claimed the top spot on Artificial Analysis's Speech Arena. The model supports 16 languages and features the ability to control its speaking style via natural language or tags like `[angry]`.

Speed, however, remains a challenge. Its generation rate of 16 characters per second reportedly looks weak compared with competitors such as Sonic 3.5 and Simba 3.2. While it leads on quality, hurdles remain for real-time use.

Halliday G2: Smart Glasses That Ditch the Camera

According to WIRED, Halliday's new smart glasses, the G2, come without a camera and are purpose-built for listening to meetings and summarizing them. With workplace meeting assistance as the primary use case, the approach appears to be a deliberate choice to sidestep privacy concerns.

With cameras long presumed to be a "must-have feature" of smart glasses, the decision to leave one out is seen as aimed at enterprise adoption. Specific release timing and pricing are covered in detail in the WIRED article itself.

Protest Against Sundar Pichai at Stanford's Graduation Ceremony

Last month, at Stanford University's commencement, more than 100 students staged a protest by walking out of CEO Sundar Pichai's keynote address. In a WIRED interview, organizers Amanda Campos and Eva Jones explained that the protest was aimed at Google's military contracts and its dealings with ICE (US Immigration and Customs Enforcement).

Whether Google should provide AI technology for defense and public safety has long been a subject of repeated debate inside and outside the company, and a protest at a public event like a graduation ceremony suggests that the relationship between young engineers and Big Tech is being shaken once again.

Community Watch: GPU Prices Whipsawed by the Falcon Exploit, LLM Distribution via Torrents, and Spreading AI Spam

This week saw several interesting developments converge in the local LLM community.

CMP 170HX and the Falcon Exploit**: A jailbreak technique called the "Falcon Exploit" targeting NVIDIA's mining GPU, the CMP 170HX, has been making headlines, whipsawing prices on the Alibaba and eBay used markets. There are also reports of sellers canceling after a deal was agreed, only to relist the item at double the price.

llama.garden**: An attempt to distribute LLMs via torrents and reduce dependence on the Hugging Face CDN has drawn attention on Reddit. The project is reportedly preparing to handle the open-weights release of Kimi K3 as well.

The spread of AI spam**: Reports are mounting of a surge in comments and spam that look unnaturally polished — apparently LLM-generated — on HN and Product Hunt. According to one poster, many of the accounts that "liked" their Show HN post turned out to be LLM bots.