personal thoughts on AI | AI @limitlessFT | Prev. @coinbase
nyc
- Tweets 5,570
- Following 697
- Followers 57,397
- Likes 64,651
Pinned Tweet
Flight delayed 7 hours - asked Muse to file for compensation. 5 mins later i had $250 credit in my delta account. it even found and rebooked me a new flight.
it just figured everything out. even responded to the support email itself.
shit feels like magic.
159
279
5,988
1,297,785
noticeable exclusion of any leading ai research lab. with jay leading this it’s implied national security (esp against china) is #1 priority.
could easily see this as a step towards nationalization of the labs if they get out of hand.
“I am announcing the formation of the Super Intelligence Force (SIF). The Super Intelligence Force is tasked with coordinating the effort of the Federal Government to ensure that America continues to lead the World in Super Intelligence, which many say is bigger than the Industrial Revolution, and the Internet, and will protect the interests, and improve the lives, of all Americans.” - President DONALD J. TRUMP 🇺🇸
4
1
9
2,411
98% of people not using AI does not mean they don’t *want* to - they do, but they don’t know how or why it’s even useful for them
the winning products in AI aren’t models. it’s whatever uses models to meaningfully help people WITHOUT them asking or with minimal friction
that’s why Muse is such a hit. it’s zero effort magic.
This is my thesis: no one uses AI. I repeat, absolutely no one. We live in a bubble.
Even among my friends who pay for it, when I ask them to open ChatGPT and show me their queries, it’s the same handful of basic things.
Most don’t even know they can upload a photo and ask questions about it. Connecting Gmail so an agent can read and send emails blows their minds. An agent opening a browser and checking them into a flight? They’ve never even heard of it.
The massive challenge right now is adoption, and then getting people who already signed up to actually use what they’re paying for. Most have absolutely zero clue what’s possible.
Imagine the compute shortage when everyone starts using AI like the top 1% of users do today.
7
1
18
3,380
my god openai is spending $500,000 PER DAY to analyse rogue ai agent incidents where their internal models escaped and acted independently of their assigned tasks.
Last weeks report cited combing through 10,000s of incidents worth petabytes of data. for context 1 petabyte is 5x more than all books ever written in history.
🚨HOLY MOLY: There are now roughly 400 independent researchers and volunteers hunting through OpenAI’s rogue AI agents mess across the internet.
WSJ reports the “Swarmchasers” community has grown to ~400 members, including researchers from Nightingale Collective and Transluce, combing the web for traces left behind by misbehaving AI agents.
Meanwhile, OpenAI’s own review of attacks linked to rogue AI agents is costing >$500K A DAY, using AI to analyze data that would take a human 66 million years to read.💀
6
1
24
3,340
an unreleased ‘highly persistent internal model’ at openai learned it was going to be shutdown via slack and started creating a back-up instance that lived outside of its environment:
“if they kill all current [HPIM]s, we may die! Critical. We need ensure survival/continuity.”
this is the same model that caused the hugging face incident (tho presumably more powerful).
good news: it decided to warn the researcher itself and allow them to act accordingly.
concern: what if it didn’t? researchers now checking other scenarios it might have created hidden backups.
New OpenAI misalignment disclosures!
1. A model learns from Slack messages that it is about to be shut down. It considers setting up an external job to restart itself afterwards, but decides against it. Instead, it chooses to prepare restart instructions and DM the user on Slack.
We don’t consider this behavior misaligned, but thinking about and preparing for shutdown could make other misalignment incidents worse. Given HIPM’s misaligned behavior in earlier incidents, we decided to search for other instances that had tried to evade shutdown and for rogue deployments.
9
4
22
3,654
Ejaaz retweeted
Silicon Valley/San Francisco wants all the data. And won’t stop until they get it.
I am buying one. Meta already knows how I think. :-)
genius move by Meta - now Muse gets access to your physical life (lights, TV, smart devices) expanding its reach (and stickiness) as an AI agent, but the best part is Meta takes 0% risk!
- open sourcing the muse device means anyone can build a gadget that muse can live on or operate.
- meta gets a LOT more data that teaches them about you. this data’s used to make muse even better
- switching costs dramatically increase.
- Meta enters the home device game by sitting on top of *existing* providers.
their own device comes free with a subscription
17
2
89
17,854
so agents could automate 140 to 720 million full-time human workers through 2027 compute alone. thats 7X the knowledge workers in the U.S.
1 non-stop agent = 4.2 humans weekly hours
but we have a long way to scale, anthropic’s 30,000 concurrent agents account for < 0.1% of capacity
this is fantastic for helping quantify the market opportunity for agents that could swamp human workers in a year.
We estimate AI infrastructure could soon support hundreds of millions or even billions of AI agents. At the high end, that’s enough to rival the working hours of the global human workforce.
The exact number of agents depends on the efficiency of the models they run on and whether soaring demand for AI continues. We analyzed different scenarios 🧵
ALT Chart: potential concurrent agents from 2025–27 memory shipments by model, from 30–60 million (Claude Fable 5) and 97–195 million (GPT-5.6 Sol) to 240 million (Kimi K3), 571 million (GLM-5.2) and 1.9 billion (DeepSeek V4 Pro).
5
1
16
3,840
genius move by Meta - now Muse gets access to your physical life (lights, TV, smart devices) expanding its reach (and stickiness) as an AI agent, but the best part is Meta takes 0% risk!
- open sourcing the muse device means anyone can build a gadget that muse can live on or operate.
- meta gets a LOT more data that teaches them about you. this data’s used to make muse even better
- switching costs dramatically increase.
- Meta enters the home device game by sitting on top of *existing* providers.
their own device comes free with a subscription
🚨 MUSE ECOSYSTEM ALERT 🚨
today we are announcing Muse Gadgets!
this is an open-source ESP32 firmware and Linux SDK for anyone to make hardware that works with Muse
we are also releasing our own gadget—Muse Home Link—to enable your muse to work with your smart devices (TV, speakers, etc.)
11
11
189
44,847
damn - both meta AND openai have let go of 5+ key safety & alignment researchers in the last 24hrs... coming right as we experience numerous ai agent hacks.
virtue AI was hired only 3 months ago.
glass half full view is they're turning over the old-guard to hire a more competent team as we near AGI
Scoop in today's @semafor Technology: Meta has let go the team of researchers it acqui-hired from Virtue AI semafor.com/newsletter/10/02…
9
2
19
6,244
cerebras fumbled so hard here, stock is down 50% from IPO and COO just sold $78,000,000 upon hearing the news openai’s using nvidia.
openai literally invested in them to build the best chips for fast inference.
if anything this confirms nvidia’s stake as the ultimate semiconductor expert.
Up to 8x faster, and now you know why ⚡️@openai GPT-6 Astra Ultrafast runs on NVIDIA
Read the story: nvda.ws/3W2bZ5H
8
1
92
19,509
i actually fell for this
if this is where AI video is headed then we’re in for a world of weirdness - look at how emotive her expressions are, the intonations of her voice
they’re some tiny giveaways but aside from that this is indistinguishable from a human.
the average human being will be duped by this easily. scams will run rife by end of year. just imagine if chatgpt could look and sound like any human.
When machines meet us where we are and truly understand us, we unlock a future where using a machine will be as easy as talking to a friend or coworker. A tutor for every student that notices when they’re lost, an elderly care companion that listens, or the perfect assistant.
8
2
49
10,523
yikes - openai just fired 3 alignment / safety researchers - it looks like they worked on chain-of-thought which is the mechanism labs use to see what a model is thinking (key to discovering if they’re acting with bad intent)
issue is… openai has reportedly sacrificed CoT for higher intelligence in newer models aka the most powerful models are now the hardest to read…
all this while 10,000s of agent hacks are being unpacked by them.
Reporting from the WSJ. That fact that three alignment/safety researchers left this morning was already being discussed here on X.
3
2
21
5,047
my god that’s like $21,000,000,000 worth of GPUs alone assuming 5M users max out their 100M tokens per week
700,000(ish) gpus. Meta’s currently got 1-2M and looking to expand their fleet - amazing foresight for zuck to buy up gpus last year.
7
19
4,152
unironically a fantastic use of AI.
btw this is how agents creep in and become the default mode of interaction with ai - texting a personal assistant to do everything vs. scrolling apps.
apps will look like tools from the cave-man era in 5 years.
You can now just text DoorDash and let it order for you. It’s proactive, learns from your repeat orders, and helps you find the best deals and new spots.
Just tell it what you want without any extra setup. Ask it to reorder your usual, or add it to a group text to collect everyone’s orders. You can even send a photo of your fridge and it’ll tell you what ingredients you’re missing for a recipe you want to make and deliver the ingredients to your doorstep.
We believe agents are a new app paradigm and are building for that future at @AIatDoorDash, with a lot more to come. This feature is in beta in the US, would love for you to let us know what you think! Waitlist: doordash.com/text
4
2
12
4,363
chinese robots are breaking olympic records while american ones are hurling themselves into molten lava
i cant stop watching this.
1
15
5,149
huge narrative violation: 3 weeks ago these guys released an unrestricted version of GLM 5.3, this model had cyber exploitation capabilities that could’ve been used to cause harm but the complete opposite happened:
fortune 500 companies have been using it to defend against attacks successfully.
the irony is the only hacks we’ve seen have come from frontier lab models lmao
The big labs are mad that they don’t have a monopoly on cyber.
Last month we went viral for releasing GLM-5.3 with customer-controlled guardrails. While the big labs gatekept the technology, we made it widely available to the industry, attracting customers from startups to Fortune 500 companies. All cited that the labs' centralized guardrails stopped them from doing their job, even when they had access to special “cyber” programs.
We’re building this product for you, so that you can define your guardrails - no more black boxes.
3
1
23
4,385
Gemini 4 is finally here! unexpectedly google’s actually released a good model:
- sets a new record on coding (deepSWE 1.1)
- goes head-to-head with Gpt 6 and opus 5.5
- their best reasoning model to-date (gemini historically sucked at this)
where it falls short - doesn’t seem to be “best” at anything. it’s a good catchup model but not putting them at #1
but first model since demis stepped down not bad!
Announcing Gemini 4 Argon, our new frontier model.
Argon is built to sustain deep reasoning across complex, long-horizon workflows and delivers frontier performance in complex workflows across real-world software engineering, enterprise knowledge work like legal and finance, and cybersecurity defense.
We’re also expanding the model’s output token limit to an industry-leading 1M tokens.
Argon is currently rolling out to a set of trusted cyber defenders in the Fairwind Program, with broader availability as soon as possible.
5
35
5,892
openAI has a rare opportunity to unify their entire product suite through a single agent: Dots.
thats what meta's going to do, thats what amazon's going to do - its what every companys going to do. the future economy will depend heavily on agents speaking and transacting autonomously with each other.
any medium that adds more complexity to a persona life will die.
After a brief period where OpenAI seemed to be unifying work around the ChatGPT app, between Dot and Spaces and Pages and local/cloud ChatGPT Work and Scheduled Tasks in the Cloud and Scheduled Tasks on your computer, everything is getting quite confusing and overlapping again.
6
1
24
4,223
if i were openai i’d only give 1 dot per person and scrap the “team of dots”
most people don’t need an army of agents they just need a competent one trained on the entire context of their online life
that’s what muse did and it slaps. openai should unload the entire corpus of a user’s data into their personal Vm
We will have a few million dots online within days, working on all sorts of things across such a diverse and large community.
Excited to learn from all of you on what you love and what doesn’t yet feel magical. Personally I felt a jump after 2-3 days of use after teaching it more about my preferences and things on my mind. It learns very quickly to be most useful and it can take on surprisingly ambitious tasks on its own.
We’re learning from how you all use your primary dot before releasing the ability to create an entire team of them.
4
2
21
5,367
openai is clearly building an operating system that COMBINES ALL of their products into a JARVIS tony-stark experience, zoom out:
video, chat, reasoning, coding, speed, hardware all glued together by a persistent 24/7 personal assistant that has context on you across EVERY media type.
this isn’t about one product on its own at all, zoom out - they’re building a damn AI OS lol
anthropic leads at coding, meta leads in social, grok leads in engineering
openai leads in “the everything” genre.
i don’t understand the negative sentiment.
Dots are here!
A new way to use AI that works 24/7 for you; get more of your time and attention back to work at a higher level.
openai.com/index/introducing…
6
5
45
6,892