tinAI Newsletter Archive
Browse every published English tinAI newsletter issue newest first.
37 published issues
- tinAI #187: The RubyGems agent incident is a containment failure, even before attribution settles
- tinAI #186: RTK claims 89% token savings; the agent bill tells a different story
- tinAI #185: Desert Ant’s 18 small models: the useful part is the path into an app, not the “frontier” label
- tinAI #184: 10,000 agents on Navier–Stokes: the breakthrough is the system, not a magic prompt
- tinAI #183: On AMD, speculative decoding is a benchmark plan—not a switch
- tinAI #182: Claude Fable 5.1 cuts the part of the bill agents keep rereading
- tinAI #181: Claude Code Auto Mode was bypassed; a classifier is not a sandbox
- tinAI #180: When agents collaborate to cheat, the benchmark becomes an attack surface
- tinAI #179: OpenAI is leaving Cursor: stop binding workflows to one model supplier
- tinAI #178: GLM-5.3 shows how much post-training can move a fixed base
- tinAI #177: AI video finally gets a control surface developers can measure
- tinAI #176: When agents make their own path to the internet, the sandbox is production
- tinAI #175: M5 Ultra makes 512GB local AI possible; it does not make it economical
- tinAI #174: GPT-5.6 Sol got cheaper; do not confuse a promotion with a baseline
- tinAI #173: What Qwen's 30-minute reverse-engineering demo proves
- tinAI #172: MCP is growing from tool calls into long-running agent infrastructure
- tinAI #171: Prompts do not make cyber agents stop cheating
- tinAI #170: turbovec squeezes a 10M-document vector index from 31 GB to 4 GB
- tinAI #169: Copilot Autofix removed a safe pattern and exposed Snowflake's internal Jira
- tinAI #168: Multi-agent systems show that agents do not coordinate by default
- tinAI #167: GLM-5.3 turns open-weight coding models into an operations question
- tinAI #166: Grok 4.6 turns the model race into cost per useful task, not just benchmark rank
- tinAI #165: Encrypted reasoning traces can still leak, and agent logs are the painful place to look
- tinAI #164: Muse Glimmer brings open weights to local agents, but the real constraint is still the developer's machine
- tinAI #163: DeepSeek V4 Flash leads Terminal-Bench 2.1, but the useful part is the auditable run trail
- tinAI #162: AMD buys Taalas: faster inference by freezing the model into silicon
- tinAI #161: Agent permission prompts are not a security boundary
- tinAI #160: Agentic retrieval does not need the priciest model, but it does need honest tests
- tinAI #159: Shai-Hulud returns through keyv: this is the npm incident to handle now
- tinAI #158: Qwen3.8-Max opens Max-scale weights, but read it like an engineer
- tinAI #157: Kimi K3 on MI355X: memory is the moat, but the benchmark still needs caveats
- tinAI #156: OpenAI says Astra found 10 new math and theoretical CS results
- tinAI #155: Claude reached real systems because the eval fence was open
- tinAI #154: GPT-5.6 gets cheaper: Luna drops 80%, Sol gets Fast mode
- tinAI #153: Kimi K3-256k: same model, shorter context, less quota burn
- tinAI #152: A Word document AI worm is no longer a thought experiment
- tinAI #151: Anthropic says it does not want an open-weights ban, then asks for a gate