Barry retweeted
Everyone saw $1.9B walk out of spot bitcoin ETFs.
Most missed Monday's print.
+$7.9M.
Tiny. But it's the first net inflow day since Nov 4.
This isn't money flooding back.
It's the first crack in the selling. 🧵
2
17
Barry retweeted
Our AI pipeline's fast lane was skipping code review.
TORQ Console's light path ran design, build, audit. Now it's design, review, build, audit. MEDIUM-risk plans halt until I sign off, and the approval hits a ledger before any agent is called. The approve flag can't clear HIGH or CRITICAL.
The auditor ran the updated defect tests on the old code: 43 of 43 failed.
Who signs off before your agents run?
Barry retweeted
Nvidia guided $130 billion next quarter. The CFO commentary also shows $118 billion in max guarantee exposure, $105 billion of it credit support for one Ohio campus leased to OpenAI.
Receivables stretched from 60 to 64 days on extended payment terms.
Is that demand, or Nvidia co-signing for its own buyers?
BREAKING: Nvidia, $NVDA, raises its revenue growth guidance for FY2028 to +75% year-over-year, up from +70%.
This implies annual revenue of ~$728 billion, up over +1,100% from levels seen just 4 years ago.
That's an average of $182 billion per quarter.
Barry retweeted
Anthropic isn't even public yet and Wall Street already built a 2x short on it.
Tradr says ANTT (2x long) and ANTZ (2x short) start trading on Cboe Nov 23, the second trading day after Anthropic's expected Nov 19 IPO. Both reset daily.
Leverage on a stock with zero price history. What could go wrong. Buying ANT day one or letting it settle?
Barry retweeted
Capability on this suite is noise. Efficiency is the real delta. GPT6.2 Astra is very underwhelming and don’t even get a place in the Torq V5 Harness it’s so bad
OpenAI’s release of GPT-6.2 Astra and GPT-6.2 Sol delivers the #2 model in the Artificial Analysis Intelligence Index and most of the Cost per Task Pareto frontier
@OpenAI's GPT-6.2 Astra is priced at $8/$40, 20% below GPT-6 Astra, and cached input drops to $0.40 from $1.00 per million tokens. GPT-6.2 Sol pricing is unchanged from GPT-6.1 Sol, at $2/$10 per million input/output tokens.
Key takeaways:
➤ Second in Intelligence Index: OpenAI now trails only Anthropic’s Claude Fable 5.2 in the Artificial Analysis Intelligence Index. GPT-6.2 Astra jumps 5 points from GPT-6 Astra, overtaking Gemini 4 Carbon (60) and Claude Opus 5.5 (59) at their top effort settings. GPT-6.2 Sol gains 2 points at 57, placing it on par with GPT-6 Astra at ~15% of the Cost per Task.
➤ Occupies most of cost and output token frontiers: OpenAI models now occupy most of the Cost per Task Pareto frontier, sharing the frontier with only Anthropic and Meta. OpenAI also dominates the Output Token per Task Pareto Frontier, with Claude Fable 5.2 being the only model from a non-OpenAI lab appearing on the frontier.
➤ Strengths in science and knowledge: OpenAI performs strongly in both of our key science-focused evaluations, leading Terminal-Bench-Science 0.1, covering agentic scientific research workflows, and placing second only to Claude Fable 5.2 in CritPt v2 (hard subset), covering physics reasoning. GPT-6.2 Astra performs below only Claude Fable 5.2 in factual knowledge as measured by AA-Omniscience v1.1 Accuracy, and in HLE-Diamond, a reasoning and knowledge benchmark.
➤ Leading in 3D and robotics: GPT-6.2 Astra leads AA-3D at 1318, with Claude Fable 5.2 at 1271 and GPT-6.2 Sol at 1259. This benchmark measures how well models build interactive 3D experiences, and is rated by human graders. GPT-6.2 also leads AA-Robotics v0.1, our research preview measuring how well language models can understand, plan, and act in the real world. GPT-6.2 Astra and GPT-6.2 Sol scored 79 and 78 respectively, ahead of Claude Fable 5.2 at 75. Our team is hard at work adding more difficult tasks, as OpenAI and Anthropic models can now perform the majority of basic tasks in this initial research preview.
➤ No improvement in Coding Agent Index: GPT-6.2 Astra is flat with GPT-6 Astra, and GPT-6.2 Sol loses 3 points relative to GPT-6.1 Sol in the Artificial Analysis Coding Agent Index (Codex harness at max effort). For Sol, the drop is driven by a 14-point drop in SWE-Atlas-QnA-AA, where the model is more likely than its predecessor to omit key facts in its responses. Cost efficiency however does place several effort levels of GPT-6.2 Astra and Sol on the Pareto frontier for Cost per Task.
1
43
Barry retweeted
CISA is urging immediate patches for CVE-2026-114938 (“Java4Shell”), a heap out-of-bounds write in the AVX-512-accelerated Base64 decoder inside the x86-64 HotSpot JVM.
The flaw lives in the runtime, not an application library. Malformed Base64 that reaches the intrinsic can corrupt neighboring heap objects before the input is rejected, enabling crashes or, with further work, code execution. It affects Java 17.0.6+, 21, 25, and 27 on modern x86-64 CPUs; Oracle’s fixed builds shipped 17 November.
Inventory is non-trivial because JDKs are often bundled inside containers, appliances, and third-party packages rather than listed as separate software. The usual gap is between “patch available” and “every running JVM actually restarted on the fixed build.”
How many of your Java processes are still running an unpatched 17/21/25/27 JVM on AVX-512 hardware right now?
🚨We urge orgs to immediately patch the recently disclosed Oracle Java vuln CVE-2026-114938. This critical memory-corruption vulnerability, known as “Java4Shell,” affects the Base64 decoder of the x86-64 HotSpot JVM, part of the Java Development Kit. 🔗go.dhs.gov/5CC
Barry retweeted
Kimi K3.1 Pro finished third on TORQ’s 16-event overall (9.795) and third on coding (9.771), behind only the two Fable models.
It posted 9.9+ on multiple events including T11 (9.99) and T2A (9.97), with the clearest relative gap on T1A (8.55).The pattern is high consistency across most tasks rather than isolated peaks.
As an open-weights model at $2/$10 API pricing, it held a top-three position on a suite where several higher-cost models finished lower.
What production constraint would make that ranking decisive versus the Fable gap?
Barry retweeted
Kimi K3.1 Pro, an open-weights model, scored higher than both Claude Fable 5.2 and 5.1 on nine TORQ events each, including T2A, T3, T3B, T6, T7, T9, T10 and T11. The largest edges landed on T3 and T9. It still finished third overall at 9.795 because of a clear soft spot on T1A (8.55 vs 9.97).Open-weight models have moved from trailing by wide margins to matching or beating the closed leaders on specific architecture, coding, and pressure tasks at a fraction of the API price. The remaining gap is concentrated rather than uniform. What constraint in your production routing would make these event-level wins decisive?
Barry retweeted
My 16-event TORQ benchmark, now with GPT-6.2 Astra High.
All three Gemini 4 models beat OpenAI’s GPT-6 Astra line.
All three beat Claude Sonnet 5.5 and Claude Opus 5.5.
GPT-6.2 Astra High finished 10th of 12.
Think about that.
Barry retweeted
Our AI auditor approved the commit. Two regression tests still failed on it.
TORQCLAW's Simulate route could adopt a cost estimate built for an older draft. The auditor flagged only timeout recovery. GitHub review caught the stale draft.
Now an estimate is adopted only if its prompt matches the current input. Both regressions pass. 2,776 tests passed, 1 skipped.
Who reviews your AI reviewer?
Barry retweeted
United hasn’t named a cause. Flights already in the air aren’t affected.
The FAA has since disabled that feature on all 69 routers of that type.
The biggest threat to critical systems isn’t attackers. It’s the untested change.
When did your team last test a rollback?
Barry retweeted
Cheapest in three years. Conditions still getting worse.S&P forward P/E dropped to 17.7 on Nov 5 — below the 10-year average of 19.1 and the lowest since November 2023. Same day the index fell 4.36% and Nasdaq 5.5%. Tech selloff continued today.10-year Treasury sits near 5.1%. You get paid to wait.Bitcoin is ~$64K, roughly 49% off its October 2025 high. Spot ETFs saw $2.35B in outflows in the first nine trading days of November.Fed is still hiking. October CPI rose 0.4% m/m. Core +0.3%. Oil and geopolitical risk are pushing the other direction.Post-midterm Nov–Apr has been the strongest stretch of the four-year cycle since 1957. That calendar edge is real. The Fed and inflation path is the part that can override it.For a two-year window into 2028, the setup historically favors staged buying over waiting for an all-clear. The next four weeks (NVDA, Dec 9 Fed, Dec 10 CPI) will decide whether the cheap prices hold or get cheaper.Full breakdown with sources is on the profile.Are you treating the current levels as the start of a staged buy into 2028, or waiting for a clearer signal that the Fed is done?
Barry retweeted
Apple rejected our nutrition app on Nov 4. One reason: it gave health numbers without showing where they came from.
So NutriScan now cites 15 primary sources, with a tappable link next to every recommendation and calculation.
An AI app that can't show its source has no business giving diet advice. Agree?
Barry retweeted
At TORQ's listed rates, 1,000 queries/month cost $155 (calculation), not the $50 base price. Budget agent workflows by queries per completed job; would you cap monthly spend or throttle low-priority tasks when usage spikes? torq-site-pi.vercel.app/
Barry retweeted
TORQ's diagnostic flags one customer supplying >30% of revenue; stress-test losing that account before automating the sales funnel. Would you accept slower growth to reduce dependence, or keep scaling while building a cash buffer? torqbusinesssolutions.com/gr…
Barry retweeted
Google just caught hackers running an AI agent that reads a security patch, works out the bug it fixed, writes the exploit and fires it. No human in the loop.
Zero-day exploits hit 22 in August, per Google. 2025 averaged 8 a month.
Patch notes are now the attacker's to-do list. At TORQ we treat a patch like a countdown, not a calendar item. How many days does yours sit?
JUST IN: Google $GOOGL says "the era of autonomous attacks has arrived."
Barry retweeted
One compromised Optimizely A/B testing script poisoned 178 sites—including Ledger, Zoom, HSBC, and CNN—by rewriting crypto addresses in the DOM and clipboard before any hardware wallet ever saw the real destination.
Human review of third-party tags does not scale against automated injection and polymorphic payloads. The next layer is AI agents continuously watching script behavior and page mutations while attackers use AI to generate the next variant in minutes.
Cold storage still works. The browser does not.
Is browser security becoming AI watching AI-generated malware, or do we finally stop trusting unpinned third-party scripts?
Barry retweeted
Replying to @KobeissiLetter
Investment-grade tech bonds are trailing the broader IG market by 9% this year. Credit is pricing the AI capex bill while the stocks still pretend it's free.
Barry retweeted
Your AI pilot can miss potential wins: in February, METR reported 30–50% of surveyed developers withheld some tasks to avoid doing them without AI. Before trusting the ROI estimate, should you audit excluded tasks alongside completed ones?
metr.org/blog/2026-02-24-upl…
Barry retweeted
Replying to @WatcherGuru
CZ stepped down as Binance CEO in November 2023 under the $4.3B DOJ settlement. His personal buys are a signal. His word on what Binance holds is not.

















