Decoding what's actually happening in AI, what's real, what's hype, what matters. The story behind the headlines.

Joined February 2018
I've been using AI every day for 2 years. These are the exact prompts I use for Perplexity, Claude, OpenClaw, Claude Code and Grok. Copy. Paste. Use today. Full free guide in the comments 👇
7
1
5
1,115
Max Decodes AI retweeted
OpenAI Codex, Day 16 of 28: Tibo reset everyone's limits, and people are already out again! Last night, Tibo: "We're cooked" This morning: "Nevermind, we're still cooking... and reset." By afternoon: usage drained, 429s, "model at capacity" Reset #6 in 15 days. ✅2 🟡3 🔧2 💎1 🔄6 🚫1 12 days left to ship something real.
6
5
36
4,191
Max Decodes AI retweeted
Has Opus 5.5 been quietly NERFED?? I reran the exact same Mentos simulation prompt from September 30 today. Same model. Same prompt. 20 days apart. Look at the difference. 💀 It looks like making the current model worse is the easiest way to make the next release look 10x better?
261
67
57
2,123
318,973
Max Decodes AI retweeted
Is Fable 5.5 coming today?
4
4
21
2,147
Max Decodes AI retweeted
MiniMax M3.1 Flash vs. DeepSeek V4.1 Flash 1017 Welllll, you get what you pay for 🤷‍♀️ MiniMax M3.1 Flash: $0 and the better Bowser DeepSeek V4.1 Flash: $0.06 and somehow worse
8
4
133
14,146
Max Decodes AI retweeted
🚨 Grok Bot can rebuild your entire X profile like a $10,000 growth consultant. For FREE. Here are 8 prompts to fix your bio, header, pinned post, and everything people see before they hit Follow 👇
1
4
7
649
Max Decodes AI retweeted
ChatGPT turns a photo of your fridge into recipes with a fully interactive cooking UI right inside the chat. One photo. One prompt. And you can actually cook along with it.
3
4
22
1,955
Max Decodes AI retweeted
Three mystery models just showed up in OpenAI's API: aurora-red, aurora-green and aurora-blue. No docs. No pricing. The last "Aurora Alpha" was a fast coding model. My bet: a Codex speed tier before the 28 days are up.
3
4
11
800
Max Decodes AI retweeted
So, does DeepSeek’s new V4.1 Flash outperform Astra 6? I don’t think so.
10
4
29
4,154
Max Decodes AI retweeted
Not canceling ChatGPT just so Claude can write me a prompt for ChatGPT 😭
1
7
1
102
30,096
You gotta be kidding me
We're extending Claude Fable 5 access on all paid plans, as well as keeping Claude Code’s weekly rate limits 50% higher, through July 19.
42
The AI hardware war just got real. Apple isn’t suing over a product. It’s suing over the people who know how to build one.
NEWS: Apple just sued OpenAI, claiming the company stole its trade secrets “at every level”, from technical staff to the Chief Hardware Officer, to build competing consumer hardware. Involved ex-Apple employees now working at OpenAI on AI hardware development.
1
43
Zuck is spending $135B on AI this year, and even he’s telling staff the agents aren’t landing like expected. When the guy throwing the most money at AI says it’s slower than the hype, maybe listen.
3
24
Sonnet 5 picked the one AI direction worth taking right now 👇
3
24
You’re about to overpay for Opus because “flagship” makes people act irrational. Meanwhile Sonnet 5 is sitting there doing near-flagship work for a lot less. Right now: Opus 4.8: $5 in / $25 out Sonnet 5: $3 in / $15 out Sonnet 5 until Aug 31: $2 in / $10 out So yes, a lot of people are about to pay extra for the badge. For agents, coding, and research, start with Sonnet 5 at high effort. Keep Opus for the really nasty judgment calls. The “just use the biggest model” crowd is not being smart. They’re just being expensive. Pick for the task. Not for the flex. That $2 / $10 pricing disappears after Aug 31.
3
29
The US government just proved it can switch off a frontier AI model. For 19 days, it did. One jailbreak found by Amazon researchers, Anthropic's own biggest investor, and Fable 5 went dark worldwide on June 12. Back July 1. The break is now blocked 99%+ of the time. Users got 50% extra usage as the apology. Bottom line: benchmarks decide if a model is good. Washington decides if it ships. #fable5
3
69