join.dev
AI news · 10 stories
llama.cpp releases version 1.0 libraries for local AI · Ai2 releases XYBench to test misguided user requests · Claude Meetings enters beta for live team collaboration
ggml-org announced llama.cpp library version 1.0.0 on November 13, with major-version bumps marking future API and ABI incompatibilities. The policy gives developers building local AI applications clearer signals about breaking changes.
Read at Hugging Face
Ai2 introduced XYBench on November 13 with more than 8,000 questions testing how assistants handle requests based on misconceptions. The benchmark evaluates whether models can address the underlying goal instead of simply following a mistaken approach.
Read at Hugging FaceAnthropic began rolling out Claude Meetings on November 12 for Team and Enterprise plans, initially supporting Google Meet. Claude can take notes, answer questions and build shared artifacts during calls, with permission controls for connected tools.
Read at Anthropic
MIT reported on November 13 a system that identifies regions of algorithmically generated structures for engineer feedback during design. Its demonstrations improved manufacturability and buckling performance, suggesting a more practical route to using topology optimization in engineering.
Read at MIT News
GitHub announced on November 13 that organizations can enable or disable repository AI Scan through its REST API at both organization and repository levels. The controls let security teams automate scan configuration across their repositories.
Read at GitHub
EU leaders gathered in Dublin on November 13 for their first AI summit, focusing on competitiveness and funding. Disagreement over future budget spending complicates plans for AI gigafactories, highlighting Europe’s infrastructure challenge.
Read at Reuters
OpenAI announced on November 12 a $2 million commitment to a scientific network studying AI and children’s safety after signing a charter at the Paris Peace Forum. The work supports shared measurement tools, with research teams retaining control of conclusions and publishing negative or inconclusive findings.
Read at OpenAI
Microsoft published commissioned IDC research on November 12 drawing on more than 5,200 business and IT leaders. Only 61% of surveyed organizations reported high confidence in measuring AI’s business value, underscoring the difficulty of assessing returns as deployment expands.
Read at Microsoft
A November 12 HFS Research study with Rackspace found that 33% of 502 surveyed AI decision-makers said their main AI program met or exceeded its business case. The survey of AI-active enterprises highlights fragmented accountability and governance gaps; it does not measure market-wide adoption.
Read at Rackspace
ElevenLabs previewed ElevenAI at its November 11 New York summit and opened free early access. Powered by Scribe v3, the app combines dictation, consent-based meeting notes and live assistance, extending the company’s speech tools into employee workflows.
Read at ElevenLabs