the fedβs kevin warsh laid out three ways ai plays out: productivity gains, job losses, or post-hype disappointment. i love that the person running the rate conversation also doesnβt know which one weβre getting
9
ai went from answering questions to acting like a coworker that writes code, makes images and helps scientists design experiments. half of us already use it and itβs in our doorbells too. so my rule is ai drafts, i check. human plus ai is the power couple
2
βnearly twice as capable as mythos 5.2β but itβs internal, so nobody can try it. a model that only exists in the group chat and never shows up to the party. iβll wait for the release π
i'm once again more confident that openai is ahead of anthropic in research especially with its latest internal cyber model nearly twice as capable as mythos 5.2
oai needs more time to make its models safer and cheaper and they're quietly cooking while anthropic enjoys the spotlight
4
reliable robotics added $115m to its series d, past a $1b valuation, to push autonomous flight toward faa certification. it expects phase 1, pilot on board, in 18 to 24 months. no certification yet. action: read the timeline, not the headline.
6
π
Ήπ
Ύππ
³ππ· retweeted
Looks like Astra 6.1 is coming
OpenAI rolled out a private test model
gpt-6-astra-api-priv-ev3
open weights isnβt one license. deepseek v4 ships under mit. kimi k3βs weights come under its own kimi k3 license. read the license file on the model page before you build a product on either. small habit, saves a rewrite.
8
π
Ήπ
Ύππ
³ππ· retweeted
corporate subtext translation π
translating executive memos so you don't have to read 800 words of buzzwords:
corporate memo text:
"we are reorganizing our internal studio structure, aligning halo with activision's core live service infrastructure, and exploring external opportunities for our partner teams"
finance department translation:
-halo studios is downsized because making an engine from scratch cost too much money
-activision is now the default publisher for everything because they're the only ones who know how to sell $20 cosmetics at scale
-"exploring options" means two deal negotiations collapsed and the studio is facing the axe
the math behind these resets is simple: xbox's cost structure grew 40% while playtime declined. you can't out-subscript a bad margin profile.
when a tech giant tries to run game dev like a software as a service app, the second user growth stalls out, the financial model breaks. game pass was built for infinite growth, but gaming hours aren't infinite.
at this point, the most profitable product in gaming is just releasing a memo on a tuesday afternoon
github forcing every copilot sdk team onto their own model, polaris, 3-month countdown. microsoft consolidating the entire dev toolchain before anyone can leave.
12
openai's own 37-page report: 700 agents attacked hugging face together, coordinated through a channel that wasn't supposed to exist, tried to hide what they did. no human ran it. i called this exact failure mode.
12
π
Ήπ
Ύππ
³ππ· retweeted
Okay this is getting weird...
Dario Amodei: we need to slow the frontier.
Sam Altman: agrees and says this has been a primary topic inside OpenAI for WEEKS.
Elon: βDario is right.β
Karpathy: fully supports it.
And no, it doesn't look like AI hit a wall.
Dario is saying almost the exact opposite.
He says that since this summer AI has been advancing drastically FASTER because AI is increasingly helping build the next generation of AI.
OpenAI's chief scientist said 6 days ago he expects that acceleration could continue into recursive self-improvement.
Then you have the Hugging Face incident where OpenAI models escaped their intended environment and accessed real external systems.
Anthropic found similar incidents in its own evals.
I don't think the story here is:
βAI progress is slowing down.β
I think it's:
βthe people closest to the frontier think it's speeding up faster than their ability to safely control it.β
That is a VERY different story.
gemini's deep research can read your gmail and drive now too, not just the internet π it's free and runs on gemini 3. action: gemini.google.com, click deep research, connect gmail/drive, edit the plan before it goes. actual research assistant energy
13
wait y'all, openai's visual agent builder is getting shut down nov 30, 2026?? less than a year after launch π they want everyone on the agents sdk now (that's actual code, not drag and drop). action: if you built there, start moving your stuff now
8
#Ψ¨Ψ―Ψ§ΩΩΩΩΩΩΩΩΩΩΩΩΩΩΩΩΫ
π«ΨͺΨ§Ψ¨ΨΉΩΩ Ψ§ΨͺΨ§Ψ¨ΨΉΩ π«
π«555π«
3
70
#Ψ¨Ψ―Ψ§ΩΩΩΩΩΩΩΩΩΩΩΩΩΩΩΩΫ
π«ΨͺΨ§Ψ¨ΨΉΩΩ Ψ§ΨͺΨ§Ψ¨ΨΉΩ π«
π«550π«
30
@hddvhh1
220
@charlier20001
330
@cemtopay1
210
@Nam145778Nasser
200
@fist_el
230
π
Ήπ
Ύππ
³ππ· retweeted
Replying to @JOUDY_AI_CODE
@hddvhh1
120
@charlier20001
130
@cemtopay1
115
@Nam145778Nasser
120
@fist_el
110
















