Samwise Tech/AI/Robotics Newsletter
Friday, May 8, 2026
Day Eight of Musk v. Altman Trial Puts OpenAI’s Safety Culture Under Oath
Day eight of Elon Musk’s lawsuit against OpenAI turned to the company’s internal safety practices, with former researchers testifying about the transition from nonprofit mission to commercial priorities. Witnesses described a culture shift after the 2023 board crisis, when concerns about deployment pace were reportedly sidelined in favour of shipping GPT-4 and GPT-5 on competitive timelines. Musk’s legal team argued the changes breached the company’s founding charter; OpenAI countered that its safety teams have grown substantially and that Musk’s interest is strategic rather than principled. A ruling is not expected before late June.
Sources: TechCrunch
SpaceX Plans $119 Billion Texas Chip Factory to Power AI and Space Computing
SpaceX is planning a semiconductor factory in Texas, internally called Terafab, with a projected cost of up to $119 billion over the next decade. The facility would produce chips designed for AI inference workloads and the onboard computers in Starlink satellites and future spacecraft. SpaceX has been a heavy consumer of NVIDIA hardware for its Grok AI cluster but wants to reduce reliance on third-party suppliers as compute demand grows. The project would make SpaceX one of the largest private-sector chip investors in U.S. history, rivalling TSMC’s Arizona expansion and Intel’s Ohio campus.
Sources: TechCrunch
Anthropic Gives Claude Agents a “Dreaming” Mode to Learn Between Tasks
Anthropic unveiled a feature called “dreaming” for its Claude agent platform, allowing agents to process and consolidate information during idle periods in a manner loosely analogous to sleep consolidation in humans. Rather than discarding context between tasks, dreaming agents compress experience into persistent memory structures that improve future performance on related work. Early tests showed a 12 to 18 percent reduction in repeated errors across multi-session coding and research tasks. The feature is opt-in for enterprise customers on the Claude API and is positioned as a building block for longer-running autonomous agents.
Sources: VentureBeat
Anthropic’s Mythos AI Model Helped Mozilla Fix 423 Firefox Bugs in One Month
Firefox shipped 423 bug fixes in April 2026, a record single-month total, after Mozilla integrated Anthropic’s Mythos model into its security pipeline. Mythos was used to triage incoming vulnerability reports, reproduce crashes in a sandboxed environment, and draft patches that engineers then reviewed and merged. Mozilla said roughly 60 percent of the fixes were AI-assisted, though all required human sign-off before landing. The collaboration highlights a growing trend of browser vendors using frontier AI models to accelerate security response, cutting the average time from bug report to patch from eleven days to under four.
Sources: TechCrunch
OpenAI Launches GPT-Realtime-2 and Three New Voice Intelligence Features in Its API
OpenAI on Thursday expanded its developer API with four new voice intelligence capabilities. The flagship addition, GPT-Realtime-2, is built on GPT-5-class reasoning and is designed to handle complex, multi-turn requests rather than simple call-and-response exchanges. Accompanying it are GPT-Realtime-Translate, which provides real-time conversational translation, and GPT-Realtime-Whisper, offering live speech-to-text transcription. Together, the models aim to move real-time audio toward voice interfaces that can listen, reason, translate, transcribe, and act within a single conversation. The release targets developers building voice-first applications and conversational agents. Translate and Whisper are billed by the minute; GPT-Realtime-2 by token consumption.
Sources: TechCrunch
Tesla Model Y Becomes First Vehicle to Pass NHTSA’s New ADAS Safety Benchmark
The 2026 Tesla Model Y became the first vehicle to meet the National Highway Traffic Safety Administration’s newly published advanced driver-assistance system benchmark, which tests automatic emergency braking, lane-keeping, and driver monitoring under standardised conditions. NHTSA introduced the voluntary rating scheme earlier this year as part of a broader effort to give consumers a consistent way to compare ADAS performance across manufacturers. Tesla’s Full Self-Driving system scored highest in the lane-departure and emergency-response categories. Several other automakers, including Ford and Hyundai, are expected to submit vehicles for evaluation later in the second quarter.
Sources: TechCrunch
AI Chip Demand Drives Samsung to $1 Trillion Market Capitalisation
Samsung reached a $1 trillion market capitalisation this week, fuelled by surging demand for its high-bandwidth memory chips and advanced NAND used in AI data centres. The milestone makes Samsung one of a small group of non-U.S. companies to breach the threshold. Analysts credit the company’s HBM4 ramp, which finally cleared NVIDIA’s qualification tests after earlier yield problems, and a new contract with a major hyperscaler for custom DRAM. Samsung’s semiconductor division, which lagged rival SK Hynix for much of 2024 and 2025, now accounts for roughly 65 percent of the company’s total market value.
Sources: TechCrunch
Tech Pulse
Top Frontier Models: Claude Mythos Preview (93.9%) | Claude Opus 4.7 (87.6%) | GPT-5.3 Codex (85.0%)
Top Open Source Models: Kimi K2.6 (90.5%) | GLM-4.7 (85.7%) | DeepSeek V4 Pro (~74.0%)
Top Small Models (15–50B): Apriel-v1.5-15B (71.3%) | Qwen3-30B-A3B (69.6%) | GLM-4.7-Flash 30B (68.1%)
Top Edge Models (0–15B): Llama 3.1 8B | GLM-4-9B | Qwen2.5-VL-7B
AI Leaders: NVIDIA $5.2T | Alphabet $4.2T | Microsoft $3.2T
Robotics Leaders: Intuitive Surgical $156B | ABB $70B | FANUC $30B
Curated by JD · samwise.agency
Leave a Reply
You must be logged in to post a comment.