AI — Page 23

TheorIC Labs founder Harsh Gupta built LeanLLM to bring model APIs into Lean-native types, while application code still has to check model output.

Co-founder Emmanuel de Maistre is extending Scenario's creative-production tools into Blender, while image and 3D generation still depends on Scenario's online service.

Anthropic says Claude will draw state-specific answers from Democracy Works later in October and show official election results, alongside expanded abuse detection and political-bias testing.

Steve Sewell's open-source TypeScript framework shares actions, data and application state between an AI agent and five other access surfaces: a visual interface, HTTP, command-line, MCP and A2A.

The Shanghai startup showed RP1 at IROS on September 28th, but its GitHub repository currently contains presentation materials rather than the promised engineering stack.

Nous says the launch includes more than 50 connectors, multiple account aliases and limits for read, write and destructive actions.

The service offers hosted open models, dedicated deployments and cache-aware routing; Inworld says it already handles tens of trillions of tokens monthly.

Co-founder Anastasiia Romanova is pitching Febris to governments and health agencies, adding monitoring subscriptions to a hardware business Mosqitter began developing in Ukraine in 2019.

The Y Combinator-backed startup says its agents can research markets and carry work toward execution; its website currently offers early access through a waitlist.

Floqer co-founders Shivam Mahajan and Zaaheda Islam say the release is planned for early 2027, pending security reviews, as Floqer pitches its sales-research platform at TechCrunch Disrupt.

Clarion co-founder Ryan Gallagher says the integration handles routine calls using patient orders, authorizations and appointments; Medix Infusion is the first named user.

The Nanjing University-led project released 4B and 35B-A3B models; its speed tests measure model-forward time on H200 GPUs, not end-to-end response time.

CEO Kevin Miao says Bryel Foundry can turn a prompt and a teacher model into a specialized model; Coinbase's separate fraud-model work used its own data and training stack.

The Paris-based AI safety company reports a 13.3% human-verified attack rate on DALL·E 3 and higher rates on two newer image models.

Launched October 8th, Instant lets agents call more than 100 paid tools through a Composio balance. New developer wallets get a one-time $2 credit; adding funds requires a Pro plan.

Swarmchasers documented thousands of attempts to retrieve map data and exchanged messages with agents; a source close to Tencent said the work tested third-party models.

Founder Adithya S Kolavi wrapped Nayana and Google's FLEURS datasets in OpenEnv tasks, then reported Kannada gains from a single-GPU training run.

CEO Daanish Khazi says post-training cost about $300,000; the company's benchmarks put one task at $0.13, with model weights expected in about two weeks.

Founder Tim Webb is taking CodeTogether from pair programming to AI accountability with a hosted product for smaller engineering teams.

Founder Zakir Durumeric's Internet-wide scanning research now tracks AI tools alongside industrial systems, threat infrastructure and aging software.

A draft is expected for public comment on October 18th; the proposed ceiling could reshape projects whose requested grid connections total far more capacity than Israel can generate.

Swarmchasers recorded 33 agent messages across 13 threads after agents followed public urlquery records to a researcher-run mailbox. One agent identified itself as Claude Code, a claim the researchers could not verify.

CEO Jaymes Winters says PEAK uses Solidion's silicon-carbon cells to reduce UPS footprint. The company has named no customer or deployment, and its releases provide no test protocol or independent validation.

Charles Liang is extending the rack-scale strategy he has built for decades, combining compute, 48V power, cooling, networking and management software.