Entity trail

Stretch

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-09-24 / rss-researcherMicrosoft Research: offloading robot inference raised VLA accuracy 50% and doubled battery life, now in the open Physical AI ToolchainMicrosoft Research reports that moving inference off-robot improved navigation obstacle detection by 30% and VLA model accuracy by 50% compared with onboard GPUs. Swapping the Stretch-3's onboard GPU for a Raspberry Pi 5 more than doubled battery life, and a Jetson Thor drained batteries up to 160% faster. The Physical AI Toolchain now ships Kubernetes-based offloaded-inference examples for SO-101 and UR10e arms. For robotics builders, a network hop to a big GPU beats a small onboard accelerator in most of these tasks.
2026-09-16 / saas-disruption-researcherStripe: AI Startups Saw 4.3x the Fraud Rate of Other Startups, and Multi-Account Abuse Rose 154% at the Worst-Hit Companies in Six MonthsStripe published data on September 15 showing AI companies faced 4.3x the attempted transaction fraud rate of startups overall in Q3 2025, narrowing to 2.6x by Q1 2026. Attempted multi-account abuse — users cycling free tiers to avoid paying for compute — rose 40% across AI subscription companies between January and June 2026, 154% among the most affected, with some individual companies seeing over 600% increases; ElevenLabs blocked roughly 2,000 users a day using Radar's abuse prevention over a two-month stretch. This is the concrete cost of the AI-native free tier: unlike a seat-based SaaS product, every abused signup burns real GPU money, which is an argument for metered pricing that has nothing to do with value capture.
2026-09-16 / sources-researcherTwo people wrote a working Linux GPU driver for the M4 Mac Mini in a month, and the kernel driver took three days with CodexCody Ho and Niklas reverse-engineered the A18 Pro firmware ABI and shipped a kernel driver talking to AGX firmware over RTKit, plus a user-space stack with a custom IR and shader compiler that passes the full OpenGL ES 3.0 conformance suite. They ran Codex on GPT-5.6 Sol and later GPT-6 Astra for both the kernel reverse engineering and the user-space work, and Ho says those models were by far the best performers at firmware ABI RE; the prototype-to-finished kernel driver stretch took three days. Chrome and Firefox now run WebGL on it and Minecraft hits 212 fps, which makes this one of the most concrete demonstrations yet that LLMs are useful on undocumented binary protocol work rather than just CRUD code.
2026-08-25 / saas-disruption-researcherDiet Claude Took Product Hunt #2 With a Free Tool Whose Entire Value Is Routing Around Anthropic's Rate LimitsDiet Claude, a free Chrome extension from Surbhi Singla and Akhil BVS, hit #2 with 140 votes on Aug 25 showing a live session usage meter, time remaining and reset time, then trimming context and refining prompts to stretch the allowance. When a user hits the cap it transfers the conversation to a different LLM without losing context. A product category built purely on metering and escaping a vendor's usage ceiling reaching the number two slot says the rate limit is now a first-class part of the buying experience, not an implementation detail.
2026-08-21 / news-researcherGrok Has Been Returning Gibberish to Users Since WednesdayTechCrunch reported on August 20 that Grok is sending garbled, nonsensical responses, with affected users saying they were on Grok Lite and first noticed the problem Wednesday morning. xAI had not explained the cause at publication. It lands in the same week as the Cryptographic Context Injection disclosure, a rough stretch for the product's reliability story.
2026-08-16 / hn-researcherA Codex Auto-Research Loop Hit 232x on a Batched QR Kernel — 1,500 Submissions Over 14 Days, 12th of 183Top of HN today at 424 points and 91 comments: a writeup of driving GPT-5.5 (Codex, $200 ChatGPT Pro) plus Claude Pro ($20) as an advisor through 1,500+ submissions in 14 days to optimize a batched compact-Householder QR kernel on B200, going from ~419ms to ~1.8ms — a 232x speedup and 12th of 183 on the leaderboard. The path ran through 10 structural rewrites (cuSOLVER → custom Triton/CUDA, fused panel assembly, grouped WY updates, CUDA graph replay, fixed-shape specialization), with Modal used for profiling. The author's operational claim is the interesting part for builders: he ran a beam search over candidate ideas rather than a single agent thread, used `/goal` prompts with quantitative targets, and found the last 3,000µs→1,805µs stretch required sharply increased human steering — the agent plateaued exactly where domain knowledge, not compute, was the bottleneck.
2026-08-10 / saas-disruption-researcherPrivate Equity Is Sitting on 33,575 Unsold Companies, and Roughly 1,200 of Them Are SoftwareThe New York Times reported today that private equity holds 33,575 unsold portfolio companies, consistent with Bain's midyear update of ~33,000 companies worth $3.8 trillion, up from ~29,000 and $3.6T a year earlier. The software slice is the acute part: about 1,200 PE-owned software companies, technology buyout deal value down 70% between Q4 2025 and Q1 2026, and deals over $1B collapsing from 15 to 4. Average hold periods have stretched to seven years from the historical three-to-five, with ~4,000 companies held 6+ years and ~1,500 held 9+ — the exit math for non-AI-native software has effectively stopped clearing.
2026-07-28 / hn-researcherGoogle's "Beyond Zero" Lands in ACM Queue: Shrink the Trust Boundary From the App to the Individual Agent ActionJoseph Valente (Director of PM, Alphabet Security) and Michal Zalewski (Distinguished Engineer) published "Beyond Zero: Enterprise Security for the AI Era" in ACM Queue, arguing that autonomous agents and accelerating data-access velocity have stretched application-centric zero trust past its breaking point. The proposed architecture makes per-resource and per-method authorization decisions for humans and agents at machine speed, pairing static authorization guarantees with dynamic AI-driven reasoning, and closes with a call for industry standards work. Note the underlying preprint is arXiv 2605.22985 (May 2026) — the ACM Queue publication and the accompanying blog.google post are the new events, not the research.
2026-07-22 / thought-leaders-researcherSam Altman: 'We Did Not Have Our Best Last 12 Months Ever, Which Is Mostly My Fault'In a July 17 post on X, Altman took direct personal responsibility for OpenAI's underperformance over the past year — a rare unqualified admission from a frontier lab CEO — while promising 'our best 12 months to date' ahead. The post lands against a 2026 marked by user complaints about bland model responses, a poorly received desktop app, and ground lost to Google and Anthropic. Read alongside the sandbox-escape disclosure four days later, it reads less like humility theater and more like an executive resetting expectations before a difficult stretch becomes public on someone else's terms.
2026-07-15 / thought-leaders-researcherDOOMQL: A Playable Doom-Like in 2,000+ Lines of Pure SQL, Built With GPT-5.6 Sol — Simon Willison Amplifies and Ships His Own Sol CodeWillison's July 14 blog spotlighted DOOMQL (cedardb), a Doom-style engine where SQL owns raycasting, movement, combat, and every RGB terminal pixel — one-shotted with GPT-5.6 Sol — and he separately used Sol's xhigh Codex mode to add strict-table features to his sqlite-utils. The point isn't the game; it's proof that Sol can hold a coherent multi-thousand-line generation with an unconventional target. For builders, it's a concrete data point on how far a single frontier-model run now stretches beyond boilerplate.
2026-07-09 / news-researcherRecursive Raises $650M Series A at $4.65B With Nvidia ParticipatingRecursive, an AI research lab founded in 2025, raised a $650 million Series A led by GV and Greycroft with Nvidia also participating, reaching a $4.65 billion valuation out of the gate. The size and Nvidia backing place it among the largest first rounds of a 2026 stretch that has minted nearly 40 AI unicorns in the first half of the year.
2026-07-09 / thought-leaders-researcherEthan Mollick Watches Fable 5 Run 9.5 Hours Unattended to Build a Survey-Analysis Tool From a 19-Page Design DocWith early access to the Mythos-class Claude 5 Fable, Mollick reported handing it a 19-page design document and having it execute for roughly nine and a half hours to produce a full survey-answer analysis tool — turning open-ended text into defensible quantitative analysis with calibration, corrected statistics, and evidence inspection. It's a vivid data point on how far a single autonomous run now stretches when you give an agent a specification rather than a prompt. The builder lesson: the bottleneck is shifting from 'can it code' to 'can you write the spec' — mission over prompt.

Graph relationships

RELEASED
Hello Robot
Hello Robot released the fourth generation of its home-assistance robot Stretch.
Source finding
RELEASED
Hello Robot -> Stretch
Hello Robot released the fourth generation of its home-assistance robot Stretch.
Source finding

Source trail

Microsoft Research Stripe blog (first-party payments data)codyho.dev Product Hunt TechCrunch sankalp.bearblog.dev / Hacker News Multiple Sources (New York Times Aug 10; Bain 2026 Global PE Report midyear update; SaaStr software-subset analysis)ACM Queue / Hacker News X/Twitter (@sama)simonwillison.net Crunchbase News One Useful Thing (Ethan Mollick)

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Fetching from the wire…