Tworek says he wants "to try and explore types of research that are hard to do at OpenAI." That sounds like a not-so-subtle dig at CEO Sam Altman's relentless focus on products and revenue, which has reportedly been causing tension among researchers. No word yet on where Tworek is headed next.
The Technology Innovation Institute (TII) from Abu Dhabi has released Falcon H1R 7B, a compact reasoning language model with 7 billion parameters. TII says the model matches the performance of competitors two to seven times larger across various benchmarks, though as always, benchmark scores only loosely correlate with real-world performance, especially for smaller models. Falcon H1R 7B uses a hybrid Transformer-Mamba architecture, which lets it process data faster than comparable models.
Falcon H1R 7B scores 49.5 percent across four benchmarks, outperforming larger models like Qwen3 32B (46.2 percent) and Nemotron H 47B Reasoning (43.5 percent). | Image: Technology Innovation Institute (TII)
Almost 90 percent of ChatGPT's roughly 900 million weekly users live outside the USA and Canada, according to The Information, citing data from tracking platform Sensor Tower. This creates a challenge for OpenAI's planned advertising business, since international users generate far less revenue. At Pinterest, for example, the average revenue per user in the USA is $7.64, compared to just 21 cents elsewhere.
India and Brazil rank among the largest ChatGPT markets alongside the USA, Japan, and France. Only about five percent of users pay for subscriptions. For emerging markets like India, OpenAI offers the cheaper "ChatGPT Go" plan at around $5 per month.
"The exponential continues until it doesn't," says Anthropic President Daniela Amodei, quoting her colleagues. At Anthropic, the team believed every year that this pace couldn't possibly keep up, and yet it did, Amodei says in an interview with CNBC TV. But that's not guaranteed, she adds. Anthropic doesn't know the future either and could be wrong about this assumption.
Economically, things get more complicated, Amodei says (from 15:56). Even if the models keep improving, rolling them out in companies can stall for "human reasons": change management takes time, procurement processes move slowly, and specific use cases often remain unclear. The key question for whether AI is in a bubble comes down to whether the economy can absorb the technology as fast as it's advancing, she suggests.
Tech companies building AI data centers are facing growing opposition from US communities, the Los Angeles Times reports. Between April and June, 20 projects worth $98 billion were blocked or delayed across eleven states: two-thirds of all projects it is tracking, according to Data Center Watch. Residents cite rising electricity costs, water consumption, noise, and loss of farmland.
Real estate firms are now considering selling land over approval concerns. Matthews, North Carolina Mayor John Higdon told the LA Times that politicians backing these projects risk getting voted out. In Indiana alone, more than a dozen projects failed to get permits, says activist Bryce Gustafson. Industry representatives like Dan Diorio from the Data Center Coalition are pushing for better community outreach.
Google engineer says Claude Code built in one hour what her team spent a year on
A senior Google engineer publicly praises Anthropic’s Claude Code: the tool built in one hour what her team spent a year developing. The quality and efficiency gains exceed anything anyone could have imagined, she says. Plus: Claude Code’s creator shares his best workflow tips.
