Samwise Tech/AI/Robotics Newsletter
Saturday, April 25, 2026
DeepSeek Releases V4 Flash and Pro, the Largest Open-Weight AI Models Yet
Chinese AI firm DeepSeek on April 24 released preview versions of V4 Flash and V4 Pro, the largest open-weight AI models available by parameter count, at 284 billion and 1.6 trillion parameters respectively, with 49 billion parameters active in the Pro variant at inference time. Both models carry a 1-million-token context window and introduce a technique called Hybrid Attention Architecture designed to improve long-context memory. DeepSeek says the models have nearly closed the performance gap with leading closed-weight models on reasoning benchmarks. Both are open-sourced under the MIT license, with the Flash model priced at $0.14 per million input tokens.
Sources: TechCrunch
OpenAI Releases GPT-5.5, Pushing ChatGPT Toward Autonomous Super-App
Less than two months after releasing GPT-5.4, OpenAI on April 23 released GPT-5.5, internally codenamed Spud. The model targets complex, autonomous, multi-step tasks: writing and debugging code, conducting online research, analyzing data, creating documents, and operating software interfaces end-to-end without constant user intervention. OpenAI says GPT-5.5 delivers improved intelligence without sacrificing speed, matching GPT-5.4's per-token latency while significantly outperforming it on capability benchmarks. The release positions ChatGPT more explicitly as a super app capable of displacing standalone productivity tools. GPT-5.5 is now rolling out to Plus, Pro, Business, and Enterprise ChatGPT subscribers.
Sources: TechCrunch
Cohere Acquires Germany's Aleph Alpha in $20 Billion Sovereign AI Deal
On April 24, Canadian AI firm Cohere announced it would acquire Germany's Aleph Alpha, creating what both companies describe as a transatlantic AI powerhouse targeting governments and regulated industries in Europe and North America. The deal values the combined company at $20 billion following a concurrent Series E, with German retail group Schwarz Group investing $600 million. Cohere shareholders will receive roughly 90 percent of the merged entity and Aleph Alpha shareholders 10 percent. Both companies have pivoted from general-purpose model development toward sovereign AI — systems built for clients demanding strict data residency and regulatory compliance, including defense, finance, and healthcare sectors.
Sources: TechCrunch
Sony's AI Robot Ace Defeats Professional Table Tennis Players in Nature Study
Sony AI published research in Nature on April 23 describing Project Ace, its autonomous table tennis robot, marking the first time a robotic system has reached competitive parity with elite human athletes in an organized sport. Ace uses 700 Hz vision sensors — ten times faster than a human eye's tracking speed — to process ball trajectory and plan return shots in real time. In match testing under International Table Tennis Federation rules, Ace defeated all three professional players it faced at least once in March 2026 contests. Sony describes the breakthrough as addressing a core challenge in physical AI: mastering high-speed, contact-rich physical environments.
Sources: Nature
Meta Signs Deal for Millions of Amazon Graviton CPUs to Power AI Agents
Meta has signed a deal with Amazon Web Services to deploy millions of Graviton CPUs to power its expanding AI agent infrastructure, Amazon announced on April 24. While GPUs dominate AI model training, the rise of agentic AI systems — which must continuously reason, coordinate, and execute multi-step tasks — is driving demand for a different compute class: faster data movement and lower latency rather than peak floating-point throughput. AWS Graviton chips, ARM-based CPUs optimized for bandwidth and efficiency, are emerging as the preferred choice for AI inference at scale. The deal signals a broader architectural shift as AI transitions from training-heavy workloads to runtime agent operation.
Sources: TechCrunch
Hospitals Are Deploying AI at Scale With No Proof It Helps Patients
An investigation by MIT Technology Review published April 24 found that despite widespread hospital adoption of AI tools — covering clinical notetaking, patient record analysis, medical imaging interpretation, and treatment recommendations — no reliable evidence yet shows these systems improve patient outcomes. Doctors are deploying AI scribes and diagnostic flagging tools at scale, but the rigorous clinical trials needed to confirm patient benefit are largely absent. The report raises questions about whether regulators and hospital administrators are moving too fast, deploying AI broadly before establishing whether it makes patients healthier or simply makes care more operationally efficient for healthcare providers.
Sources: MIT Technology Review
AI-Powered Astronomy Is Worsening the Global GPU Crunch
Astronomers using AI to analyze telescope data are contributing to the global GPU shortage, according to a TechCrunch report published April 23. With NASA's Nancy Grace Roman Space Telescope scheduled to launch in September 2026 — eight months ahead of schedule — the observatory is expected to generate over 20,000 terabytes of data during its operational life, requiring AI systems to process and classify it at scale. Astrophysicists competing for GPU cluster access say commercial AI training runs are increasingly crowding out scientific workloads. The convergence of big-data astronomy and AI is intensifying pressure on already strained GPU supply chains worldwide.
Sources: TechCrunch
Tech Pulse
Top Models: Claude Mythos Preview (93.9%) | Claude Opus 4.7 (87.6%) | GPT-5.3 Codex (85%)
Market: NVIDIA $4.3T | Alphabet $3.89T | Apple $3.81T
Curated by JD · samwise.agency
Leave a Reply
You must be logged in to post a comment.