All Stories — Page 8
Every story published on RuntimeWire, newest first.
- Dynamic Creatures introduces Hugo, a mobile robot built for guest interactions
Marc Theermann's startup says Hugo connects perception to movement through its Vision-Language-Expression model, targeting theme parks and hospitality venues.
- RuntimeWire flagged DeepSeek's V4.1-Flash-1017 refresh 3 hours before its announcement
RuntimeWire published its report at 12:08 a.m. CT on October 17th; DeepSeek posted the official release at 3:35 a.m. CT, about 3 hours 28 minutes later.
- Anthropic gives six METR staffers internal access to assess frontier AI risks
Announced October 15th, the arrangement lets the nonprofit publish risk assessments with limited redactions. A separate POLITICO report published the next day described Anthropic pitching lawmakers on a draft AI safety bill, but did not date the pitch.
- Neuralwatt says cache misses can use up to 140% more inference energy
CEO Chad Gibson's analysis of Neuralwatt Cloud traffic links prompt edits to wasted GPU work, with the biggest costs in long agent sessions.
- Anduril executive says Steelfin could eventually be made in Europe
Brian Moran told Reuters Anduril would consider European partners for Steelfin; flight tests are due in 2027.
- Hark offers $1M in free pizza to demonstrate AI ordering
Brett Adcock's U.S.-only promotion routes users through Hark Pro and DoorDash, turning a routine food order into a live test of its consumer AI.
- FineEnvs reports Kannada OCR gains after selecting a checkpoint on test data
FineEnvs says it replaced its test-selected OCR checkpoint with the final training checkpoint on Oct. 13th; current public records identify step 500. The team also reviewed its ASR test curve.
- Replit CEO says it stopped working with Anthropic, citing cost and model tests
At TechCrunch Disrupt, Amjad Masad praised OpenAI models. Replit's September production notes and June Claude integration leave unclear what he means by 'stopped working.'
- NVIDIA opens ASPIRE, a robotics agent that stores skills as code
Jim Fan's team says the system can refine robot-control programs and reuse successful skills, with reported benchmark results reaching 99%.
- Nuno Maduro and two PHP developers built /dev/null in 48 hours
The October 15th-16th sprint produced /dev/null, a GitHub-login social feed for developers that its creators kept tuning in public.
- ChatGPT adds an iOS widget for one-tap access to its Dots agents
The widget opens a Dot conversation or starts a user-initiated voice call; Dots launched in September for a limited set of paid plans.
- Michael Denyer syncs 15 pstack commits into a seven-runtime agent port
GitHub lists pstack-claude v0.9.86 as released on October 15th, 2026, with a verified signature. The update pins upstream pstack v0.15.25, adds two skills and records 15 sync conflicts resolved by hand.
- Hyperscale Data targets an askROI listing without choosing a deal
The subsidiary could pursue an IPO, SPAC or reverse takeover in the second or third quarter of 2027; founder Todd Ault is expected to hand leadership to askROI executives.
- Elorian launches Glimpse to give AI a second look at visual evidence
Andrew Dai's first model for the startup revisits images and video as it reasons, with company-reported benchmark wins and clear gaps against frontier systems.
- CarbonAI raises $12M as Kaufman points to $1T in CRE refinancing
Kaufman says limited lender review capacity is driving demand for automation; CarbonAI reports shorter screening time for one customer.
- OpenAI adds parallel agents to GPT-6 Pro in ChatGPT
The Chat tab rollout lets Pro users track delegated tasks, but OpenAI's post gives no latency figures or examples to support its claims that GPT-6 Pro responds faster and uses plugins and tools better.
- Pika brings its creative toolkit into ChatGPT
The integration adds Pika's image, video and sound tools to ChatGPT, with a side panel, timeline editing and access to users' Pika libraries.
- Claude traces NIST's ECDSA curve seeds to 1990s timestamps
Anthropic says Claude identified the generator behind five prime-curve seeds, while several curve parameters remain unexplained.
- Mercor estimates 90% of US work hours lack AI benchmark coverage
Brendan Foody's company mapped 29 benchmarks to Labor Department job tasks, exposing a gap that also supports Mercor's push into AI evaluation.
- China proposes export controls for frontier AI model technology
A draft would require pre-export registration for listed technologies, including model weights meeting specified thresholds; public-domain AI technology is proposed for exclusion.
- Calif demonstrates an untethered jailbreak on the iPhone 18 Pro
The Hexacon demo restored root access after a reboot on iOS 27.0.1; Calif has not disclosed the vulnerabilities or released a tool.
- Veritas adapts its anti-counterfeit chip for jewelry, trophies and perfume
Veritas founder Luci Holland's second-generation chip works on metal and through glass, with ATHLOS set to put it in 18 championship crowns.
- RoboParty opens RP1's robot designs, control stack and training code
Founder Yi Huang built robots in his Harbin dorm before starting RoboParty; RP1's public materials show both the reach and current limits of its open-source bet.
- China proposes export controls on 40 technologies across AI, biotech and solar
The ministry says the draft covers technology and certain equipment; exporters must already register proposed transfers while the ministries gather comments through November 15th.