• OpenAI introduced Dots as always-on agents that learn users’ working preferences and can take on tasks from routine follow-ups to multistep coding and testing, using connected tools and a cloud computer with customizable instructions and safeguards. OpenAI ChatGPT Space adds shared pages where teams can work with Dots through comments and task handoffs; Altman said Dots and Space were available that day, with Dot conversations not counting toward usage. OpenAI OpenAI also described specialist enterprise Dots that teams can give goals, context, and feedback, and said internal testing found them effective in many roles. OpenAI
• Altman announced ChatGPT “6.1 Sol,” positioned as a cost-capability balance with near-Astra intelligence and cheaper context reuse for longer tasks. OpenAI OpenAI also announced UltraFast across its API, ChatGPT, and Codex, claiming output of 300 tokens per second. OpenAI
• OpenAI made Codex available in the cloud, allowing users to start a task on a phone and continue on desktop, and announced an API with the Codex harness, multi-agent controls, and computer use. OpenAI
• Altman said OpenAI had reached its previously stated goal of an AI research intern; research lead Tejal reported that models were completing over a third of day-long research tasks without intervention by July, after previously failing most such tasks. OpenAI
• Altman argued that AI’s future should look more like a Renaissance than an Industrial Revolution: AI should give people more power in their lives, not make them “cogs in a machine.” OpenAI

• Altman argued that AI coding tools have let developers do more without reducing the amount of work: expectations and latent demand have grown. He expects AI to take on work requiring little creativity where people do not care who performs it, while demand for new things to do remains Every.
• Altman said his agent filters overnight activity to preserve his morning focus and can surface time-critical matters; a colleague’s agent also warned of a likely demo failure minutes before a presentation Every. He described it connecting context across Slack and his calendar Every, building five or six versions from rough notes Every, and finding a screenshot after Codex failed to locate it Every. He said a sufficiently capable model was a key enabler, alongside other system work Every.
• Altman described Space as an ambitious bet because existing software is not designed for close collaboration among people and multiple AIs; he highlighted live documents that update as people or circumstances change and can be worked on by multiple agents Every.
• At DevDay, he said OpenAI wants to open ChatGPT to third-party apps and a marketplace, with subscription portability, and argued that small teams and one-person companies can build significant products rather than one model company capturing everything Every.
• Altman said OpenAI aims to offer capable models across the cost, speed, and modality trade-offs, while recognizing it cannot cover every new architecture Every. He described speed as a new democratization priority and said OpenAI’s own chip would help deliver fast inference broadly; he forecast an 8×-faster option at a great price in the not-distant future, followed by 100× faster, while expecting a higher-priced frontier to persist Every.
• He said safety alignment and security had recently become major concerns for him, compared with revenue and growth around six months earlier Every.

• Hassabis said artificial general intelligence—systems capable of exhibiting all human cognitive capabilities—remains his goal after more than 30 years, and that substantial work remains to reach it. 마인드셋
• He said DeepMind released AlphaFold’s predictions openly and freely, and reported that roughly 2 million researchers across 190 countries had used it. 마인드셋 He also described collaborations with the Drugs for Neglected Diseases initiative, supplying protein structures for work on diseases including leishmaniasis and dengue to support drug design and the search for cures or mitigations. 마인드셋
• Hassabis expects many major advances over the next decade to come from intersections between disciplines; he said DeepMind’s research groups bring together fields including engineering, machine learning, neuroscience, mathematics, philosophy, ethics, chemistry and physics, and credited multidisciplinary collaboration as a strength behind work such as AlphaFold. 마인드셋
• He argued that AI could bring productivity, economic and scientific gains, while society should decide collectively how to use it and distribute its benefits, with technologists engaging civil society, governments, academia and industry. 마인드셋 He also warned about misuse by bad actors and the challenge of keeping increasingly autonomous, agent-like systems under control with appropriate goals, values and robust guardrails. 마인드셋

• LeCun argued that LLMs mainly distill existing human knowledge and produce new knowledge only in domains such as mathematics and code where their outputs can be checked; he said agentic systems remain unreliable because they cannot predict the consequences of their actions. La Tribune Événements His alternative is a world model that learns abstract, predictive representations from video or other inputs, discarding unpredictable details rather than trying to generate video pixel by pixel. La Tribune Événements
• In a six-month post-launch update, LeCun said his company was still in research, had no near-term product plan, had 50–60 people across Paris, New York, Montreal and Singapore, and was talking with potential clients and partners. La Tribune Événements He identified industrial process control as a near-term application, including fault prediction, maintenance and process optimization; he said initial releases would be research papers, then B2B systems deployed through industrial partners, with consumer robotics and autonomous vehicles further out and no firm timeline. La Tribune Événements
• Separately from his company, LeCun described Tapestry as an open foundation-model project, including LLMs, intended as an AI equivalent of Linux and built with governments, universities, companies and researchers. La Tribune Événements He said countries and regions could contribute to training a shared model without disclosing their data, retaining data ownership and sovereignty. La Tribune Événements
• For France and Europe, he urged support for open AI platforms, greater investment in compute capacity and research, better researcher compensation, and easier access to venture capital. La Tribune Événements

• OpenAI introduced “Dots,” described as always-on agents that take on delegated work using user context; examples include monitoring projects, using connected plugins and a cloud computer, and writing and testing code, with customizable instructions and safeguards. OpenAI The presenter said Dots and ChatGPT Space were available that day to ChatGPT Pro, Business, Premium, and Enterprise users; Dots were included in the plan and their conversations did not count toward usage. OpenAI ChatGPT Space is a shared workspace where people and agents can collaborate on pages and visualizations, tag agents to act, and keep work in sync; shared presentations were described as coming soon. OpenAI
• OpenAI announced a lower-cost model described as offering intelligence close to Astra, plus “UltraFast” across the API, ChatGPT, and Codex; the presentation claimed UltraFast reaches 300 tokens per second. OpenAI It also introduced a decision API that selects from predefined options to respond in a fraction of a second. OpenAI
• OpenAI research lead Tejal said the company’s internal tracking showed models completing over one-third of day-long research tasks without intervention as of July, after previously failing most such tasks. OpenAI OpenAI also said its models had helped solve more than 100 math problems open for decades and were contributing to work including antibiotic development and ancient-language research. OpenAI
• OpenAI launched Codex in the cloud and announced an API with the agent harness, multi-agent controls, and computer use. OpenAI It described a privacy offering as providing frontier-model safety without storing content on its servers, and claimed API reliability above 99% and a 45% reduction in time to first token. OpenAI
• OpenAI said developers could let users sign in to apps with ChatGPT and use tokens from their ChatGPT subscription, avoiding those costs for developers; the launch began with 16 partners. OpenAI It also announced plug-in extensions for ChatGPT and Codex, in-conversation discovery, and a simplified review process. OpenAI

• Six months after launch, AMI Labs was still focused on research, with no near-term product plan; LeCun said it had 50–60 people across Paris, New York, Montreal and Singapore and was speaking with potential customers and partners. La Tribune Événements
• LeCun argues that language models predict the next token and distill existing human knowledge, but are unreliable because they cannot anticipate the consequences of actions. His alternative is world models trained to predict real-world outcomes from video or other inputs, using abstract representations that discard unpredictable details—such as individual moving leaves—and focus on relevant dynamics such as vehicles and pedestrians. La Tribune Événements
• He sees industrial process control as a near-term use for world models, including detecting anomalies, predicting failures and maintenance needs, and improving efficiency or reducing emissions; he sees robotics as a medium-term application. La Tribune Événements
• LeCun’s Tapestry project, launched in Paris in early May, aims to build an open, freely usable foundation model with contributions from governments, universities, companies and researchers, which others could adapt into applications; he likens the shared platform to Linux. He presents open platforms as a route to AI sovereignty and a counterweight to dependence on a few US or Chinese providers. La Tribune Événements
• For France and Europe, he calls for support for open platforms, greater computing capacity—saying Europe lags the scale of US private investment—and stronger research capacity to attract leading researchers and connect them with industry. La Tribune Événements

• AMD is acquiring World Labs in a deal valued at $8.2 billion; the startup builds AI models that understand and reason about the physical world. The deal was expected to close by year-end, subject to regulatory approval. Bloomberg Technology
• Fei-Fei Li will join AMD as executive vice president and chief scientist, reporting to CEO Lisa Su; Su said Li would contribute across AMD’s roadmap. Bloomberg Technology
• Li described World Labs’ move as mission-driven: AMD had been its investor and chip provider from the beginning, and discussions with Su had developed a shared vision. Bloomberg Technology
• Li called AI safety a shared responsibility, saying AI must be developed and deployed responsibly through efforts spanning individuals, companies, industry, and society. Bloomberg Technology

• Sam Altman discussed OpenAI holding back Astra 6.1 and pausing training on another model amid what he called a steep capability curve; he said the company would slow training or model releases when needed to focus on safety, alignment, monitoring, and security, and must maintain confidence in its safety cases. Bloomberg Podcasts
• He proposed shared AI safety standards defining what is needed at each capability level to proceed without undue risk, plus ways to check that labs implement them, such as independent evaluators, government review, or companies checking one another’s work. Bloomberg Podcasts
• Altman said explaining evolving AI risks and benefits to the public is a responsibility tied to OpenAI’s mission, not to whether the company goes public. Bloomberg Podcasts

• Sam Altman said OpenAI held back a model release as capabilities enter a steep growth phase. He said the company will slow training or releases when needed to focus more on safety, alignment, monitoring, and security, while continuing to ship releases it considers reasonable, such as faster or cheaper versions. Bloomberg Podcasts
• He called for shared safety-case standards defining what is needed at each capability level to proceed without undue risk, along with shared ways to verify implementation, such as independent evaluators, government review, or companies checking one another’s work. Bloomberg Podcasts
• Altman said OpenAI aims to improve AI quality while driving prices down, arguing that better AI encourages more use and that users want speed, price, and capability. Bloomberg Podcasts

• The interview described AMD’s planned acquisition of World Labs as an $8.2 billion all-stock transaction, expected to close by year-end subject to regulatory approval. Bloomberg Podcasts Lisa Su said Fei-Fei Li would become AMD’s executive vice president and chief scientist, reporting to Su, with the World Labs team contributing across AMD’s roadmap. Bloomberg Podcasts
• Li described World Labs as a frontier model company focused on spatial and physical intelligence. Its Atlas multimodal model generates pixels with 3D/4D consistency and allows cameras and agents to interact in generated worlds; Li said the work extends beyond language modeling to needs in design, education, healthcare, and robotics. Bloomberg Podcasts
• Li said World Labs’ growing ambition also increased its need for compute, and argued that joining AMD could accelerate the hardware–software development cycle: models benefit from cutting-edge hardware, while hardware development benefits from insight into future workflows and needs. Bloomberg Podcasts
• Li argued that open ecosystems help cross-pollinate discovery, innovation, entrepreneurship, and business opportunities. Bloomberg Podcasts She also described AI safety as a shared responsibility, calling for responsible development and deployment across individual, company, industry, and societal levels. Bloomberg Podcasts

• Altman described Dots as a high-capability product aimed initially at enterprise users and prosumers, including people building startups, with a mass-market consumer version to follow; he said it uses OpenAI’s best model, Astra. Bloomberg Television
• Asked about holding back Astra 6.1 and pausing training on a separate model, Altman said OpenAI would slow training or releases when needed to focus more on safety, alignment, monitoring, and security as capabilities rise; he said faster or cheaper versions could still be released. Bloomberg Television He also called for shared safety-case standards that set requirements for proceeding at each capability level, with implementation checked through options such as independent evaluators, government review, or companies checking one another. Bloomberg Television
• Altman said OpenAI expects to go public eventually, but wants first to learn how to operate as a public company and make decisions under new safety requirements without the added pressure of being newly public; he said those transitions should not happen all at once. Bloomberg Television

• AMD is acquiring World Labs in an all-stock transaction valued at $8.2 billion; the deal was expected to close by year-end, subject to regulatory approvals. Bloomberg Television Fei-Fei Li is to become AMD executive vice president and chief scientist, reporting to Lisa Su; Su said the World Labs team would contribute across AMD’s roadmap. Bloomberg Television
• Li described World Labs as a frontier-model company focused on spatial and physical intelligence; she said its Atlas model generates pixels with 3D/4D consistency and lets cameras and agents interact within generated worlds. Bloomberg Television She said joining AMD would accelerate the software–hardware development flywheel and help address World Labs’ growing need for compute. Bloomberg Television
• On AI safety, Li called it a shared responsibility and argued that responsible development and deployment require effort at individual, company, industry, and societal levels. Bloomberg Television

• Sam Altman said Nvidia’s AI-guardrail software was a good step and that OpenAI was doing similar work, but he was not deeply familiar with its details and did not consider it a complete solution. He warned that treating AI safety only as an engineering problem would overlook the unresolved scientific challenge of discovering how to align models. CNBC Television
• On AI liability, Altman expected a multi-level framework, analogous to the auto industry, with responsibility depending on whether a failure came from the model, how it was used, or intentional misuse. CNBC Television

Sam Altman said that as models become more capable, OpenAI’s safety cases must meet higher standards, with alignment, safety, monitoring and security staying “way ahead of capabilities.” CNBC Television He characterized the model that was not launched as slightly worse on a few evaluations—not a major alarming issue—and called the decision an abundance-of-caution measure. CNBC Television Altman described pacing as improving safety and monitoring so OpenAI can make confident safety claims and continue progress responsibly, rather than simply stopping development or always slowing it down. CNBC Television

• AMD announced it would acquire Fei-Fei Li’s World Labs in an all-stock transaction; the broadcast gives conflicting values—$8.2 billion and $8.2 million—so the deal value is unclear in the transcript. Bloomberg Television
• Li described World Labs as a frontier-model company focused on spatial and physical intelligence, and said its Atlas model has 3D/4D consistency that lets cameras and agents interact within generated worlds. Bloomberg Television
• Li argued that AI models and hardware will co-evolve: advanced hardware can accelerate model training and inference, while model workflows can inform future hardware needs. AMD CEO Lisa Su said World Labs’ research team would form the nucleus of AMD’s models team, complementing its hardware, software, and systems capabilities. Bloomberg Television

• AMD’s proposed acquisition of World Labs was described as an $8.2 billion all-stock transaction Bloomberg Technology, expected to close by year-end subject to regulatory approval Bloomberg Technology. Lisa Su said Fei-Fei Li would join AMD as executive vice president and chief scientist, reporting to Su Bloomberg Technology.
• Li described World Labs’ focus as spatial and physical intelligence, and Atlas as a multimodal model with 3D/4D consistency that lets cameras and agents interact within generated worlds Bloomberg Technology. She said the AMD combination would help meet the company’s growing compute needs and accelerate the hardware-software development flywheel Bloomberg Technology.
• Li called AI safety a shared responsibility, said AI must be developed and deployed responsibly, and urged work across individual, company, industry, and societal levels Bloomberg Technology.

AMD is acquiring World Labs in an $8.2 billion all-stock transaction; Fei-Fei Li said the merger reflects a shared vision for AI and will help accelerate the hardware–software development loop as World Labs’ ambitions and compute needs grow. Bloomberg Podcasts

Li described World Labs as a frontier-model company focused on spatial and physical intelligence beyond language modeling. Its Atlas model generates images with 3D/4D consistency and allows cameras and agents to interact within generated worlds; she connected this work to needs in creation and design, education, healthcare, and robotics. Bloomberg Podcasts World Labs also expanded into robotics through its acquisition of Synapse, whose team built a robotic simulation workflow and robotic-policy-training models; Li said advanced hardware can accelerate model training, inference, and servicing, while insight into future workflows can inform hardware development. Bloomberg Podcasts

• At the All-In Conference, Cohere CEO and co-founder Aidan Gomez said he expects AI to augment work rather than simply displace it Amber Mac. He cited chess’s rising popularity despite computers outperforming humans, and said software demand and software-engineering employment had increased after LLMs began automating software engineering—contrary to an Anthropic CEO forecast that 90% of coders would be automated within 12 months Amber Mac.
• To reduce dependence on AI, Gomez recommended teaching reasoning and other fundamentals without AI first, then introducing it once learners have mastered them, like calculators after arithmetic basics Amber Mac. He also favored testing without AI models to identify learning gaps and argued that education could address risks, rather than treating declining human capability as inevitable Amber Mac.
• On Canada’s technology ambitions, Gomez argued that major social-media, consumer-device and cloud waves produced companies in the United States while Canada and other democracies failed to create comparable leaders; he urged building those capabilities and framed current geopolitical and AI shifts as a rare chance for Canada to raise its global profile Amber Mac.

• Sam Altman warned that increasingly capable, autonomous AI could outpace institutions and become difficult for people to understand or control; recursive self-improvement and automating AI development could accelerate progress further. He said competitive pressure does not justify rash risks and argued against training systems without strong evidence for alignment, monitoring, safety guarantees, and human control. The Financial Express
• He called for consequential AI decisions to be shaped through democratic processes and accountable governments, alongside international cooperation on frontier AI standards for measuring capabilities, assessing risks and safeguards, and preserving human oversight. He also proposed shared approaches to evidence and compliance, rapid incident reporting, and secure channels for sharing vulnerabilities and threats; standards should not entrench incumbents or favor one business model, and governments should choose how to implement them in their own legal systems. The Financial Express

• Yann LeCun argues that free, diverse AI platforms will be needed as AI mediates more digital interactions, to serve different languages, cultures, value systems, and interests rather than concentrating those choices in a few companies or countries. He says open-source infrastructure offers greater security and customizability and enables ecosystems, pointing to Linux as a precedent. World Governments Summit
• LeCun argues that fears of imminent AI takeover are misplaced: AI remains far from human and animal learning capabilities, and current LLMs lack physical-world understanding and persistent memory and do not reason or plan at the level desired; he says scientific breakthroughs are still needed. World Governments Summit