TL;Der — AI/ML

tlder@devAI/ML
tlder@dev:~$

Filtered to AI/ML. 82 items. covering May 1 → Jul 17

└─worth-opening/(36 items)

The clearest read on what Murati's lab is actually betting on — customization over raw benchmark wins — and why a 975B open-weight release matters right now.
└─·ai-ml
The post details how the model makes turn-taking decisions many times per second and delegates to GPT-5.5 for heavy tasks — the architecture is more interesting than the headline suggests.
└─·ai-ml,mobile

└─cross-cutting/(51 items)

AI companies have until July 22 at 18:00 CEST to sign the EU AI Office's Code of Practice and secure a presumption of regulatory conformity before Article 50 enforcement kicks in August 2.
└─·leadership,ai-ml·TechTimes
CISA added CVE-2026-55255 (Langflow authorization bypass), CVE-2026-48908 (JoomShaper SP Page Builder unrestricted file upload), and CVE-2026-56290 (Joomlack Page Builder access control flaw) to the Known Exploited Vulnerabilities catalog on July 7.
└─·ai-ml·CISA
Jul 15
Jul 15·cat news/20260715-models-inkling-open-weight
cat news/20260715-models-inkling-open-weight
Thinking Machines drops Inkling, its first model — and the weights are open
Mira Murati's lab released Inkling, a 975B-parameter mixture-of-experts model with open weights, built from scratch in under nine months.
└─TechCrunch
Jul 9
Jul 9·cat news/20260709-models-gpt56-sol-terra-luna-pub
cat news/20260709-models-gpt56-sol-terra-luna-pub
GPT-5.6 goes public after U.S. Commerce clearance
OpenAI's three-model GPT-5.6 family went public July 9 after US Department of Commerce approval: Sol tuned for science and cybersecurity, Terra matching GPT-5.5 at half the cost, Luna as the lightweight option.
└─Nextgov
Jul 8
Jul 8·cat news/20260708-models-openai-gpt-live-1-voice
cat news/20260708-models-openai-gpt-live-1-voice
OpenAI Releases GPT-Live-1 and GPT-Live-1 Mini — Full-Duplex Voice Models Now in ChatGPT
OpenAI ships two dedicated voice models that can listen and speak at the same time, replacing the current Advanced Voice Mode and landing in the API.
└─OpenAI
Jul 8
Jul 8·cat news/20260708-models-openai-devday-2026-deadline
cat news/20260708-models-openai-devday-2026-deadline
OpenAI DevDay 2026 Application Deadline Is July 10
OpenAI set July 10 as the cut-off to apply for DevDay 2026, giving interested developers two days to get their submissions in.
└─Digg / OpenAI
Jun 5
Jun 5·cat news/20260605-llms-claude-opus-41-deprecation-dead
cat news/20260605-llms-claude-opus-41-deprecation-dead
Claude Opus 4.1 Deprecated — August 5 Retirement
claude-opus-4-1-20250805 entered deprecated status on June 5 and retires August 5, 2026, with claude-opus-4-8 as the replacement.
└─Anthropic Platform Docs
Jul 9
Jul 9·cat news/20260709-llms-anthropic-claude-reflect-dash
cat news/20260709-llms-anthropic-claude-reflect-dash
Anthropic introduces Reflect — a dashboard to track how you use Claude
Anthropic launched Reflect, a beta analytics dashboard inside Claude that shows usage patterns, topic breakdowns, and AI habit insights for Free, Pro, and Max users with memory enabled.
└─Anthropic / TechCrunch
Jul 9
Jul 9·cat news/20260709-llms-anthropic-hard-questions-qa
cat news/20260709-llms-anthropic-hard-questions-qa
Anthropic asks the public for their hardest AI questions, commits to answering transparently
Anthropic launched a public Q&A initiative soliciting the most difficult questions about AI safety, promising substantive and transparent answers rather than polished non-answers.
└─Anthropic
Jul 8
Jul 8·cat news/20260708-llms-fable5-billing-now-active
cat news/20260708-llms-fable5-billing-now-active
Fable 5 Per-Token Billing Is Now Live
Anthropic's Fable 5 per-token billing activated today, July 8, completing the transition away from subscription inclusion.
└─Anthropic
Jul 7
Jul 7·cat news/20260707-llms-chinese-models-us-share
cat news/20260707-llms-chinese-models-us-share
US Developers Keep Reaching for Cheaper Chinese Models
Cost pressure is quietly pulling US developers toward cheaper Chinese models, even as export controls dominate the policy headlines.
└─Industry analysis
Jun 16
Jun 16·cat news/20260616-llms-claude-mythos-preview-retire-june30
cat news/20260616-llms-claude-mythos-preview-retire-june30
Claude Mythos Preview Retires June 30
The claude-mythos-preview model endpoint will stop accepting requests on June 30, with claude-mythos-5 as the migration target.
└─Anthropic Platform Docs
Jun 15
Jun 15·cat news/20260615-llms-claude-sonnet4-opus4-retired
cat news/20260615-llms-claude-sonnet4-opus4-retired
Claude Sonnet 4 and Opus 4 Original Builds Fully Retired from the API
As of June 15, claude-sonnet-4-20250514 and claude-opus-4-20250514 no longer accept API requests — the retirement date arrived and both are gone.
└─Anthropic Platform Docs
Jun 15
Jun 15·cat news/20260615-llms-claude-sonnet4-opus4-eol
cat news/20260615-llms-claude-sonnet4-opus4-eol
Claude Sonnet 4 and Opus 4 Retire June 15
Anthropic is retiring Claude Sonnet 4 and Opus 4 on June 15, requiring API callers to migrate to current model versions before that date.
└─Anthropic Docs
Jun 1
Jun 1·cat news/20260601-llms-anthropic-prog-billing-june15
cat news/20260601-llms-anthropic-prog-billing-june15
Anthropic splits programmatic Claude usage into a separate credit pool starting June 15
Claude programmatic usage — CI runs, Agent SDK automation, GitHub Actions, third-party agent frameworks — moves to a dedicated monthly credit pool on June 15, separate from interactive usage.
└─devtoolpicks.com
Jul 28
Jul 28·cat news/20260728-mcp-spec-rc-stateless-final
cat news/20260728-mcp-spec-rc-stateless-final
MCP 2026-07-28 Specification Goes Final — Sessions Removed, OAuth Hardened
The Model Context Protocol's largest revision since launch publishes its final spec on July 28, dropping stateful sessions entirely and adding MCP Apps, Tasks, and OAuth 2.0/OpenID Connect alignment.
└─Model Context Protocol Blog
May 12
May 12·cat news/20260512-mcp-figma-mcp-server-launch
cat news/20260512-mcp-figma-mcp-server-launch
Figma Launches Official MCP Server for Bidirectional Code-Design Workflows
Figma released an official MCP server enabling AI agents to translate code into Figma designs, modify design systems, and convert designs back to code via the open MCP standard.
└─Figma Help Center
May 5
May 5·cat news/20260505-mcp-jama-connect-mcp-server
cat news/20260505-mcp-jama-connect-mcp-server
Jama Software Launches MCP Server for Jama Connect, Enabling AI Coding Assistants to Access Requirements Traceability
Jama Software has released an MCP Server for Jama Connect, letting AI coding tools such as Claude, Codex, Cursor, and GitHub Copilot query and navigate requirements traceability data while respecting existing permissions and audit workflows.
└─IT Business Net
Jul 1
Jul 1·cat news/20260701-agents-gemini-spark-mac
cat news/20260701-agents-gemini-spark-mac
Gemini Spark reaches the Mac with local file access
Google's agentic Gemini Spark assistant arrives in the macOS app, able to work with local files and connected apps for Ultra subscribers.
└─TechCrunch
Jun 23
Jun 23·cat news/20260622-agents-openai-patch-the-planet
cat news/20260622-agents-openai-patch-the-planet
OpenAI Launches 'Patch the Planet' Open-Source Vulnerability Initiative
OpenAI, Trail of Bits, HackerOne, and Calif jointly launched Patch the Planet, an effort to find and fix vulnerabilities in widely used open-source software.
└─OpenAI
Jun 18
Jun 18·cat news/20260618-agents-hf-agentic-bench-open-mo
cat news/20260618-agents-hf-agentic-bench-open-mo
Hugging Face Publishes Benchmark Guide for Open Models on Real Agentic Tooling
A new Hugging Face blog post walks through evaluating open-weight models against your own tool definitions rather than canned benchmarks.
└─Hugging Face Blog
Jun 18
Jun 18·cat news/20260618-agents-mosaicleaks-research-agent
cat news/20260618-agents-mosaicleaks-research-agent
MosaicLeaks: Benchmarking Whether Research Agents Leak Confidential Information
ServiceNow published a benchmark via the Hugging Face blog testing whether research agents inadvertently surface information they were told to keep confidential.
└─Hugging Face Blog
Jun 18
Jun 18·cat news/20260618-frameworks-hf-beyond-lora-peft-co
cat news/20260618-frameworks-hf-beyond-lora-peft-co
Hugging Face Compares LoRA Alternatives Across PEFT Fine-Tuning Methods
A Hugging Face guide benchmarks several PEFT fine-tuning techniques against LoRA, the de-facto default, to see whether newer methods actually improve on it.
└─Hugging Face Blog
May 20
May 20·cat news/20260520-frameworks-google-ai-studio-io26
cat news/20260520-frameworks-google-ai-studio-io26
Google AI Studio Gains Native Kotlin Support, Workspace Integrations, and Cloud Run Export
Google expanded AI Studio at I/O 2026 with native Kotlin support, Google Workspace integrations, one-click Cloud Run deployment, and direct export to Antigravity.
└─Google Developers Blog
May 19
May 19·cat news/20260519-frameworks-autogpt-cve-33233-rce
cat news/20260519-frameworks-autogpt-cve-33233-rce
AutoGPT Unsafe Pickle Deserialization Enables Arbitrary Code Execution (CVE-2026-33233)
A CVSS 7.6 vulnerability in AutoGPT allows remote code execution through unsafe pickle deserialization.
└─CVE Records
May 19
May 19·cat news/20260519-frameworks-autogpt-cve-33232-dos
cat news/20260519-frameworks-autogpt-cve-33232-dos
AutoGPT Unauthenticated DoS via Disk Space Exhaustion (CVE-2026-33232)
CVE-2026-33232 allows an unauthenticated attacker to exhaust disk space on AutoGPT instances, rated CVSS 7.5.
└─CVE Records
May 1 → Jul 17AI/ML82 entries