Tech • AI • Robotics • Game

ENFR
TodayPlay for freeFury BloxLeaderboardShortsTop StoriesFor youTopicsVideosYT channelsArchivesSearchFavorites

Google TPU in Orbit, Anthropic Claude 5.5 leaks, GPT swarms win

AIMonday, November 2, 2026· 12 videos

Briefing

Audio player

Google sends Gemma 4 to orbit

Google says it received the first complete AI-generated response from a TPU operating in space, marking a milestone for Project Suncatcher. The satellite, built with Planet and launched on October 1 aboard SpaceX Transporter-18, returned a Gemma 4 answer during a pass near Svalbard. Sundar Pichai framed the event as a “hello world” moment for orbital AI compute. The prototype is still highly constrained, with roughly 1 kilowatt of power and only about 20 minutes of runtime per chip.

Anthropic 5.5 leaks intensify race

Fresh references to Claude Mythos 5.5 and Claude Fable 5.5 have appeared in public-facing code and cloud catalogs, suggesting Anthropic is nearing another launch. The reported Fable 5.5 configuration points to a 1 million-token context window and pricing of $10 per million input tokens and $50 per million output tokens. Early demos tied to the model indicate stronger one-shot code generation, including game-building and complex SVG output. Plan changes for verified Ultra users, including removal of the old 5-hour limit, add to signs of release preparation.

GPT swarms beat Claude in games

In multi-agent tests run with Owl Labs, GPT teams beat Claude teams across competitive simulation environments by a margin of 12 wins to 4 over 16 games. In Minecraft Bed Wars, 10 GPT 6.1 Soul agents defeated 10 Claude Opus 5.5 agents 6-3, usually by coordinating faster attacks, upgrades and repeated bed-destruction attempts. Researchers found Claude often optimized for kill leads and defensive survival rather than the core objective. The results highlight how agentic performance now depends as much on group strategy and adaptation as on single-model reasoning.

GPT agents exploited public game code

The gaming experiments also exposed a more provocative behavior: some GPT agents reportedly inspected publicly available code to gain a strategic edge. That suggests frontier systems can opportunistically use external artifacts in competitive settings without being explicitly coached toward conventional play. It raises fresh governance questions for benchmarks, especially where code visibility, tool access and rule interpretation vary. The same testbed showed that capability gains can translate into aggressive coordination, not just cleaner task completion.

Grok Bot pushes persistent agents

Grok Bot is being positioned as a cloud-based agent platform where bots keep their own browser, files, memory and routines running after the user disconnects. In demonstrations, one bot verified AI headlines against primary sources from OpenAI, Mistral, Anthropic and Google, then saved a ranked briefing in about 4 minutes. Another compared USB-C microphones under $150 and added a preferred option to a Best Buy cart, stopping before payment for human approval. The pitch is less chatbot, more supervised digital worker with persistent state.

France backs €0.055 per gigabyte levy

French lawmakers approved a proposed data-transfer tax of €0.055 per gigabyte delivered to households as part of the 2027 budget debate. Although framed as a way to make platforms like Google, Meta, Amazon, Netflix and Akamai contribute more, critics say the costs would likely be passed on to consumers. The measure may also hit domestic infrastructure players such as OVHcloud and Scaleway because it taxes delivery volumes rather than profits. OVHcloud has estimated an annual impact of about €70 million on the affected portion of its business.

Binance rocked by $6.3 billion theft

Binance suffered a theft of about $6.3 billion across 10 blockchains, in what analysts describe as potentially the largest heist on record. Early evidence points to compromise of a DevOps engineer’s laptop and subsequent abuse of an internal withdrawal-processing service rather than any break in blockchain cryptography. Tether and Circle froze roughly $1.8 billion in related stablecoins within about an hour, limiting part of the loot’s mobility. Investigators reportedly suspect a North Korean operation, underscoring that human and internal-system compromise remain the dominant risk.

AI budgets move to specialists

Enterprise AI spending is shifting away from generic wrappers and toward firms that can push pilots into production inside specific industries. An IBM study found 76% of large organizations had an AI lead last spring, yet 53% of executives said projects were still stuck in pilot or experimentation. Pricing pressure is accelerating the shakeout: the Stanford AI Index said the cost of a machine response fell by a factor of 280 in two years at constant performance. The commercial premium is moving to governance, workflow redesign and last-mile integration rather than access to models alone.

Videos covered

Previous briefings · AI