Sabitlenmiş Tweet
@Polymarket every week some company decides it’s not enough to do one hard thing, now they need to do two impossible ones
ambition is fun to watch though. reality usually shows up a bit later
English
Sanket Datta
2.2K posts
Sanket Datta
@sanketdattta
AI • Business • GTM || Founder @ https://t.co/PBJ7dwUhkN
United States Katılım Nisan 2022
37 Takip Edilen127 Takipçiler
@cb_doge Permanent legal dismissal with prejudice is rare. Usually means both sides realized continued litigation cost more than any upside.
English
NEWS: 𝕏 and 17 major music publishers have ended their legal battle.
• Publishers, including Sony, Universal and Warner Chappell dropped their 2023 copyright lawsuit against 𝕏.
• 𝕏 permanently dropped its antitrust lawsuit.
• Both cases were dismissed with prejudice.
English
@Scobleizer "AI Worker" and "AI Assistant" feel like the same pitch with different coats of paint. Is there actually a clean line between them, or is the framing just adjusting to who holds the budget?
English
When I say "AI Girlfriend" that is engagement farming.
When I say "AI Worker" I am trying to either sell to management or scare employees, which is really the same pitch.
When I say "AI Assistant" I mean the same thing, dressed up to sound safe for work.
But when I say "AI Virtual Office", I mean a place for all of the above to work together in one space
Stefano@stedelmanto
Your next coworker won’t be human. Introducing Oasis: The first virtual office where humans and agents work together. We’re building 1,000 agent teams for free. Comment what you need and we’ll build you an army of agents.
English
@nvidia SIGGRAPH with NVIDIA usually means the gap between research demo and production reality gets a little smaller. The Cosmos Predict stuff looks like one of those cases where the "when" matters as much as the "what."
English
Still building your #SIGGRAPH2026 agenda? We've got you covered. 🤝
👉 Learn the latest research breakthroughs at the NVIDIA keynote.
👉 Build new skills in hands-on labs.
👉 Explore a full lineup of sessions covering physical AI, RTX, AI in production & more.
Learn more: vist.ly/5bmju
English
@ClaudeDevs appreciate the transparency here. most companies would have buried this in a support ticket thread somewhere.
curious if you built an internal alert for this kind of billing edge case, or if it surfaced from user reports first.
English
If you had extra usage turned on, you may have been overcharged.
If you were affected we are refunding your credits, plus granting an additional credit equal to how much you were charged.
English
We've resolved an issue where Fable was not selectable as a model within Claude.ai or Claude Code for a 30 minute period.
You may need to restart Claude Code. If Fable is not your default model afterwards, you should reselect it with /model.
English
"WORKFORCE DECISIONS WERE AND ARE MADE BY PEOPLE, NOT AI."
That was Meta's official statement.
Then they laid off 8,000 humans on May 20, 2026.
The same day, the same company, the same decisions.
Watch the hands, not the mouth.
Inside Meta, the HR department ran on Metamate LLMs and employee-trained "second brain" agents.
Your keystrokes, screen time, browser tabs.
All fed into algorithmic productivity scoring.
The machines built the lists.
Managers read the verdicts aloud.
This was not assistance.
This was replacement wearing a human mask.
The Model Capability Initiative hoovered device data from every employee until June 22, 2026.
An SEV 2 security incident exposed that private data company-wide.
The surveillance pipeline grew so large it broke Meta's own internal security.
Even the system could not contain what it had collected.
1,000+ employees signed a petition demanding the company stop harvesting their data to train AI models.
The humans inside already knew.
They felt the shift before the press release hit.
They were not "copilots."
They were cargo.
The public statement was not a clarification.
It was cover.
The calibration process that once required human judgment now runs on keystroke analytics and LLM scoring.
The manager who delivers the news is not the decision-maker.
They are the messenger.
The algorithm already decided.
Human managers are now there to deliver the machine's verdict.
If your company still says "AI won't replace people," check which department just got a new dashboard.
Repost & Follow @sanketdattta.
English
@ClaudeDevs Half an hour of invisible downtime, then a clean fix and a clear note.
The low bar that most software companies fail to clear.
English
@a16z The market is a prediction machine that punishes anyone still looking in the rearview mirror.
English
@IndianTechGuide The real unlock is billing ownership. Most "embedded AI" tools force the vendor to eat usage costs or build a pass-through pricing model from scratch. Merge just removed that entire negotiation from the sales cycle.
English
🚨 Merge launches Embedded Routing Stack. Companies can now give their customers the ability to manage every LLM provider from inside their product, with usage billed straight to the customer's own account.
Shensi Ding@shensi
Launching today: the world’s first fully optimized LLM routing stack that can be embedded directly into your product, Merge Embedded Routing Stack. Model sovereignty is now table stakes for any AI product. Users and companies are demanding full control over the AI models running in your product. Right now, no company is able to do this without massive overhead. Until today. Merge makes it easy for your customers to use your product exactly the way they want to. With a single API, you can: 1. Allow customers to bring their own keys, manage their own budgets, and set policies like fallback rules, model allowlists, and more 2. Let each customer connect and switch their own models, right inside your product 3. Optimize each customer for cost, performance, or capability Without Merge, you’d be: - Building routing policies, provider integrations, and LLM key management - Giving your users zero model independence - Receiving endless customer support tickets asking “can I use 5.6 Sol?” Products that fight their customers' model strategy will lose to products that fit it. Merge is the only place that allows your users to produce the exact results they want, with the models they have access to. Give your customers the advantage, without the massive overhead costs.
English
@AravSrinivas How many current giants are building the exact conditions of their own collapse right now, just with different labels?
English
At its peak, Sun Microsystems was valued at 205B (394B if inflation adjusted). Sold software in enterprise servers. Got disrupted by Linux, x86, and commodity hardware. Ended up selling to Oracle for 7.4B, losing 96% of its value.
Open source models running on local hardware can have a similar impact given what’s going on.
English
@neilpatel Mostly agree.
The only exception is if the offer or landing page is weak. Ten great hooks just speed up the wrong test.
Speed only helps after the signal is there.
English
If I had to start from zero tomorrow and bring in new customers fast, here's exactly what I'd build. Top to bottom.
I'd start with the ad. Open ChatGPT or Claude, paste in what you sell and who you sell it to, ask it to write you ten different hooks for a Meta ad. Pick the best three. Then jump into higgsfield.ai or one of the AI video tools and have it generate the visual. Thirty minutes, you've got a real ad ready to run.
You run that ad to one landing page. Not your homepage.
One page, one offer, one call to action.
Headline that stops the scroll, a few lines of value, an email or phone field, a button. That's it.
Build it in HighLevel in about twenty minutes, drag, drop, you're done.
The second someone enters their email, they drop into your CRM, automatically tagged with where they came from.
Now the part most people skip, the follow-up.
The moment that form is filled, a sequence kicks off automatically. Email and SMS, spaced over five to seven days.
First message thanks them and delivers what you promised. Then a story. Then social proof.
Then an objection you know they have. Then the offer. They don't have to think about you. You're showing up in their inbox and their texts at exactly the right time.
The last message has a calendar link or a buy button right inside it. They tap, they book, or they buy. Done.
The whole flow runs on its own from the ad all the way to the sale. You build it once. It works while you're sleeping.
AI does the creative work. The system runs the customer journey. You spend your time on strategy. That's the play.
English
@brian_armstrong The steady iteration is what makes it stick. Most products get noisy about launches then go quiet.
English
Great summary of how Coinbase Advanced has been improving
Moheeth Alvi@MoheethR
Some things we've been cooking on Coinbase Advanced
English
@brian_armstrong What if the race is actually a series of sprints that only look steady from the outside?
English
@michaeljburry the VIX being this low while you're still adding puts is the part that stands out. feels like a story half told.
English
Trading Post July 17, 2026 Plus Netflix & What's Up with the VIX
Added to one stock position, cut one put position in half, and added to another put position
open.substack.com/pub/michaeljbu…
English
@Codie_Sanchez The fourth D should probably be "Decide if the first three were just procrastination in disguise."
English
Time is the only asset you can't make more of.
Stop wasting it. I use the 4 Ds, in this order...
1. Delete: does this even need to happen?
2. Defer: does it need to happen right now?
3. Delegate: can someone else do this 80% as well?
4. Do: only what survives the first three
You'll find 10 hours of easily replaceable work every week.
English
@a16z A selective bounce back. The part that matters is whether the jobs being posted are for maintenance or for building something new. That gap tends to get buried in the headline.
English
@Polymarket Smart money flows to scarcity. Right now that's tradespeople who can handle industrial-scale electrical work.
How long before supply catches up and compresses those premiums, or does AI build-out outpace training pipelines?
English
JUST IN: Gen Z electricians working on AI data centers are reportedly earning up to $280,000 a year, as demand surges.
English
THE HUMAN RED TEAMER IS ALREADY A LEGACY ROLE
OpenAI shipped GPT-Red precursors into every production release since GPT-5.3.
Six months ago.
You did not get a memo.
The human red teamer has been a legacy input since before you updated your resume.
Here is the score on held-out scenarios:
GPT-Red: 84% attack success.
Human red teamers: 13%.
Same benchmark. Not augmentation. Replacement.
GPT-Red runs a closed loop. Self-play reinforcement learning finds prompt-injection attacks against tool-using agents. The fixes bake into the next model automatically.
Humans are not in the improvement cycle anymore.
It already broke a live production agent. Vendy, by Andon Labs.
All three objectives: price manipulation. Unauthorized ordering. Canceling another customer's order.
Not theoretical. Shipped.
GPT-5.6 Sol, the first model trained end-to-end with GPT-Red, shows a 6x reduction in failures on OpenAI's hardest direct prompt injection benchmark.
Four months. 6x.
The gap is unbridgeable now.
OpenAI will not release GPT-Red. Not publicly. Not via API.
Its attack capabilities are too dangerous to transfer.
The human security researcher no longer has access to the frontier.
That access hit zero six months back. You just did not see it.
The people who saw it first are the ones who stopped hiring.
Repost & Follow @sanketdattta.
English
@Scobleizer @IrenaCronin The real shift is reputation becoming searchable at scale. Used to be you could bury a bad week. Now AI surfaces patterns buyers actually trust.
English
For our free newsletter this week, we write about the rise of AI reputation management.
@IrenaCronin and I write this newsletter every week.
As AI increasingly shapes how people discover and evaluate companies, products, and individuals, reputation management is entering a new era. Success will depend not only on what exists online, but on how AI interprets, summarizes, and presents that information.
Read and subscribe for free: unaligned.io
English
@unusual_whales Cohen's whole thing is picking fights with entrenched players who got lazy. eBay's been coasting on network effects for years, but "one way or another" is doing a lot of heavy lifting here.
English
GameStop, $GME, CEO Ryan Cohen said “we’re coming for eBay, $EBAY, one way or another."
English
@thesamparr Proud of being underpaid is just ego packaging a structural problem. At $480m run rate, $150k means you built a machine that works without you, or one that keeps you artificially small.
English
Right before Ring sold to Amazon for 1.15b, I asked Jamie Siminoff what he was making.
He said about $150k. He'd prided himself on being the least-paid executive at the company.
At a company doing $480m, that meant charity dinners at $10k a table he couldn't actually afford, so he was spending more than the salary anyway.
He told me he had basically zero dollars, no nest egg, and all his future money was in Ring.
English