b/acc🔜 @aidotEngineer CODE 🗽
@AccBalanced
Followers
9K
Following
307K
Media
8K
Statuses
225K
AI Factories. Balanced Accelerationist. WEKA CAIO, CNCF kubernetes founding board, Post-PKI.
Seat 14D
Joined July 2008
OH at @aiDotEngineer “I work on AI at @GEICO, and I wanna make AI so simple a caveman could do it”
0
0
0
Announce an upcoming Country-of-Origin feature Go silent for 2 weeks Suddenly announce that it’s launching soon Release it on everyone’s profile—but users can only see their own country Everyone is confused why it’s private So they post screenshots of the feature, helping us
1K
610
16K
Dude... Nana Banana Pro turning my CoreWeave analysis into a lecture white board.. What is this black magic.
14
31
488
GPT-5: “After 2 days of thinking…” Something tells me the AGI-2027 timeline is going to revert when those “2 days of thinking” models come out.
Yep! Things seem to be going somewhat slower than the AI 2027 scenario. Our timelines were longer than 2027 when we published and now they are a bit longer still; "around 2030, lots of uncertainty though" is what I say these days.
23
40
434
google's genius: by not selling TPUs, google allows Nvidia to maintain high GPU pricing, which in turn props up market price of inference. google then captures that inference price premium by running TPUs for inference on GCP. $goog
wait the worlds greatest ai required no $nvda...
74
294
6K
I give credit to Cloudflare for literally posting source code for what caused the outage.
84
353
17K
It’s been 48 hours with Gemini 3 in @AmpCode. The model is impressively good, but the rollout wasn’t as smooth as we wanted it to be. Here are the issues we’ve seen: 1. Capacity issues and related early stopping (now resolved, and monitoring) 2. Thinking leaks 3. Repetitive
8
6
100
Nano Banana Pro, released this morning, is clearly the best image generation model. Superb instruction following, plus it can generate full infographics (with correct spelling and properly rendered text!) from a short prompt based on running extra searches
simonwillison.net
Hot on the heels of Tuesday’s Gemini 3 Pro release, today it’s Nano Banana Pro, also known as Gemini 3 Pro Image. I’ve had a few days of preview access …
25
81
723
MiniMax on their mission & AGI > mission: intelligence for everyone, not just a few > MiniMax-M2 is 230B‑A10B for a reason > impossible triangle: performance, speed, cost, usually pick two > MiniMax-M2 breaks that triangle > near-SOTA, 2–3x faster, 8% cost of closed models >
8
18
138
remember when people thought github copilot was going to eat everything? then we got: cursor, claude code, codex, cognition, antigravity, & many many other (vibe) coding products. we are still pretty early in a lot of ai categories that people think are pretty closed off or
27
7
407
Been this way. Folks just now starting to realize how dead internet is
Social media is about to get weird 🍌 I just built an entirely new Instagram profile with the newly launched Nano Banana Pro. Not one of these photos is real. Prompt below ⬇️
50
70
991
Two revelations this week have shaken the narrative in AI : Nvidia’s earnings & this tweet about Gemini. The AI industry spent 2025 convinced that pre-training scaling laws had hit a wall. Models weren’t improving just from adding more compute during training. Then Gemini 3
12
16
60
This sounds bad: @antigravity is vulnerable to the classic lethal trifecta exfiltration attack where a prompt injection can cause the agent to construct a URL to an external server controlled by the attacker and then invisibly leak stolen data to it by rendering a Markdown image
Friendly advice, Be careful when using Google's Antigravity IDE with sensitive data (API keys, secrets...) Attackers can hide instructions in code comments, documentation pages, or MCP servers and easily exfiltrate that information to their domain using Markdown Image rendering
23
79
750
Reminder: This is a log plot. A straight line here already means exponential progress. You don’t need any "superexponential 2027" bs. This is already insane.
37
33
813
I think Aug-Dec 2025 will be remembered as the big jump in models ability to read intent. Starting with GPT-5 and continuing with Grok 4.1 and Gemini 3 (with Opus 4.5 sure to follow), they all got much better at inferring intent from text - both human intent and character/story
15
12
168
Naming tech startups after Lord of the Rings is a thing, there's a still a few good ones up for grabs. Someone should use these... Taken: - Palantir - Anduril - Durin - Mithril Available: - Gandalf - Galadriel - Elrond - Lindir - Aragorn - Rohan - Shire? Just thinking
448
72
2K
pretty unexpected, honestly… Gemini 3 Pro is still hitting an 88% hallucination rate in this eval, same as 2.5 Pro and 2.5 Flash basically no improvement in refusal or accuracy. Sonnet 4.5 and GPT-5 handle this much better because they say "no" more often, so their answers
80
114
989
Nano Banana Pro just took in the entire Nvidia Q3 earnings PDF and generated this beautiful infographic. This is the world's best compression engine.
106
164
2K