Overview
Fourteen posts drove today’s feed. Gemini 3 leaks and benchmarks, xAI vs Google scorekeeping, a guerrilla Rocky anecdote, another internet wobble, Congress pushing to release the Epstein files, Cloudflare taking half the web down, and a Roblox “awareness” game that backfired.
The big picture
Gemini 3 Pro model card leak
Tanishq Abraham posts an archived Gemini 3 Pro model card, showing a multimodal, 1M-context model beating Claude 4.5 and GPT-5 on a string of benchmarks, especially maths and code.
🔗 https://x.com/iScienceLuvr/status/1990738932951036178
ARC-AGI jump for Gemini 3
ARC Prize confirms Gemini 3 Pro and Gemini 3 Deep Think have doubled prior state of the art on ARC-AGI-2, with Deep Think at 45.14 percent but at a steep 77 dollars per task.
🔗 https://x.com/arcprize/status/1990820655411909018
One-shot 3D LEGO editor
Pietro Schirano shows Gemini 3 Pro building a full browser 3D LEGO editor in one go, plus game and emulator demos, arguing we are entering a new app-generation phase.
🔗 https://x.com/skirano/status/1990813093727789486
Google announces Gemini 3 Pro
Logan Kilpatrick launches Gemini 3 Pro as “the world’s most intelligent model”, with top scores on maths, science and tool benchmarks and a preview in AI Studio and the Gemini API.
🔗 https://x.com/OfficialLoganK/status/1990813077172822143
Rocky’s steps were guerrilla
Emir Han shares Stallone’s story that the Rocky steps run was shot on the fly without permits, just grabbing the moment before police could arrive, which fits the film’s scrappy DNA.
🔗 https://x.com/RealEmirHan/status/1990860332693070325
“Grok is the model in the arena”
Elon Musk points at Grok 4.1’s top spot on LMSYS Chatbot Arena, where it currently edges past Gemini 2.5 Pro and Claude Sonnet 4 in blind human preference tests.
🔗 https://x.com/elonmusk/status/1990661253702844692
“The future of AI is open source” intern
Okara posts a screenshot of a new intern’s first-day suggestion to tweet “the future of ai is open source”, which the community instantly approves as on-brand and hire-worthy.
🔗 https://x.com/askOkara/status/1990701115680436257
Cloudflare outage question
The Kobeissi Letter notes that a global Cloudflare outage has hit over 100 million users, after two major AWS incidents this year, and asks why outages feel more frequent.
🔗 https://x.com/KobeissiLetter/status/1990782803043385576
Cloudflare recovery meme
Watcher Guru covers Cloudflare’s recovery with a screenshot of the Android restart menu, capturing how many people first blame their phones before realising the web itself is wobbling.
🔗 https://x.com/WatcherGuru/status/1990786793344733358
House votes to release Epstein files
Thomas Massie announces his bill with Ro Khanna to release the Epstein files has passed the House 427–1, moving long-sealed documents closer to public view.
🔗 https://x.com/RepThomasMassie/status/1990870378885828773
Gabe delivering Steam Decks
Pirat Nation reminds everyone that Gabe Newell personally dropped off Steam Decks near Valve HQ in 2022, which remains one of gaming’s favourite “CEO actually turned up” stories.
🔗 https.com/Pirat_Nation/status/1990887743824490856
Google’s Windsurf fork with stray branding
Silas Alberti points out that Google’s Antigravity IDE still has “Cascade” text in places, suggesting a hurried fork of the Windsurf codebase after licensing the tech.
🔗 https://x.com/silasalberti/status/1990898984706036125
“Gifts are non-refundable”
Sir Dickson shares a DM from someone claiming a gifted amount helped them win 98 million naira on bets and asking how to repay; his caption says gifts are non-refundable and wishes them blessings.
🔗 https://x.com/Wizarab10/status/1990687653105447114
Roblox HRT game critique
Rock Solid comments on a satirical Roblox game about hormone replacement that crashes when “HRT” levels get too high, arguing the game design shoots itself in the foot and raising concerns about tone and audience.
🔗 https://x.com/ShitpostRock/status/1990770464789311655
Why it matters
Frontier model race tightens
Gemini 3 leaks, ARC-AGI scores, Google’s official launch and Grok’s Arena lead all land within a day. Benchmarks suggest Gemini 3 Deep Think can reason better on structured tasks, while xAI leans on human preference wins and lower hallucination. Different yardsticks, same arms race.
Software as conversation
The 3D LEGO editor and game demos show how “describe what you want, get a full app” is moving from demo to something closer to normal, at least for prototypes. That shifts where value sits: framing the problem and judging the result.
Infrastructure stress
Cloudflare and previous AWS outages underline how dependent the internet is on a handful of providers while AI traffic keeps rising. It only takes one bad day in one company for X, ChatGPT and half your usual sites to go missing.
Transparency and trust
The House vote on Epstein files and the Windsurf/Antigravity branding slip both speak to trust. One is about exposing sealed records, the other about how big players adopt start-up tech and how transparent that process really is.
Culture, games and identity
The Rocky anecdote, Gabe’s Steam Deck deliveries and Aaron Paul’s game work yesterday all push gaming and film nostalgia. The Roblox HRT game, and the replies around it, show how quickly games aimed at awareness can turn into flashpoints if the metaphor misses.
