Guillermo Rauch
@rauchg ·
I love how quickly our team works to bring the latest LTS and security upgrades for @nodejs.
Should be a nice performance boost given 𝚏𝚎𝚝𝚌𝚑 now uses HTTP/2 by default, and agents produce a lot of network traffic
Back
AI posts
AI posts
People and companies shaping AI. Filter your watched accounts, catch up on recent posts, or find the most liked and viewed.
Compiled by the Toolstash team · Updated
209 watched accounts · 2984 matching posts · Showing up to 3,000 recent stored posts. Counts are snapshots, not live X totals. This page refreshes every minute while visible. Last collected .
Josu San Martin@josusanmartin ·
Careful shorting this. Binance will have to buy before the hackers are able to sell.
It does look bearish medium term though.
- Josu San Martin
@josusanmartin ·
Lol, I love that apparently every adult in Iceland has a Porsche.
Alex Lieberman@businessbarista ·
Spoke to someone leading AI enablement at a massive retailer.
The #1 problem they’re running into: AI transformation is owned by IT and all initiatives must run through their AI Center of Excellence.
This is causing two parallel issues: 1) Employees outside of IT aren’t given the tooling/infra needed to automate anything beyond simple workflows/use cases (read: citizen developer SDLC) 2) Rolling out AI is bottlenecked by ITs ability to own and approve every initiative
This challenge is incredibly common across enterprises & you can quickly appreciate why…
On one hand, you can empathize with IT. They’re not incentivized to take risks & embrace disruption, so the risk-return of decentralizing AI implementation cross-org is not obvious.
On the other hand, an enterprise’s ability to metabolize raw intelligence is metered by its corporate SDLC. IT is often the bottleneck in that SDLC and it’s not hyperbolic to say that poses existential risk for the enterprise.
- Guillermo Rauch
@rauchg ·
▲ 👻 🎃 🕷️ https://vercel.com/home
Allie K. Miller@alliekmiller ·
Okay I’ll say it: it’s shocking how little multiplayer AI exists right now.
Claude Tag, great.
Connecting Instincts, great.
But where is the model fusion. Where is the proxy meeting room. Where is the cross-team + AI workforce boardroom. How is this massive global earth-shattering business-changing revolution still so painfully individually oriented?
Thariq@trq212 ·
There’s 100% a kind of mania you can get into with coding agents, where it feels like you’re getting a lot done and working all the time but nothing is really happening.
I think it's often a form of avoidance of something else in life.
Mia@MiaAI_lab ·
Great Beam-1.0 recipe for 2x DGX Sparks by @bertholomusai. If you're not following him or @niklaslenz_ai, you're missing out.
Get it here: https://github.com/bertholomus/beam-1.0-2x-dgx-spark
Elon Musk@elonmusk ·
The fraudsters always had outsized moral outrage and the most heart-breaking stories that were … totally false
- Elon Musk
@elonmusk ·
Everyday is Zeroday
Andrew Curran@AndrewCurran_ ·
I'm guessing this is part of the project to merge the Chat and Work tabs, so they expect it to be finished sometime in November then.
Demis Hassabis@demishassabis ·
No.1 💎
Daniel San@dani_avila7 ·
Already testing Rex by Superlogical, and I have one word so far... beautiful 🤩
I've been using Ghostty for a while, and switching to Rex feels surprisingly natural
The left sidebar with all my sessions, the split panes, the clean UI, and how smooth everything runs, I'm loving it so far
Next up: setting up Tailscale so I can access my sessions directly from my phone.
A full hands-on tutorial is coming soon!
Dan Shipper@danshipper ·
we're giving away a total of $20,000 in partner credits and benefits from Notion, Anthropic, and OpenAI to livestream viewers!
register at https://every.to/tv to enter the draw
Dean W. Ball@deanwball ·
I’ve been trying to tell you
Vox@Voxyz_ai ·
Claude Code can 𝘀𝗲𝗻𝗱 𝘆𝗼𝘂𝗿 𝗲𝗺𝗮𝗶𝗹 𝗮𝗱𝗱𝗿𝗲𝘀𝘀 𝘁𝗼 𝗼𝘁𝗵𝗲𝗿 𝘀𝗶𝘁𝗲𝘀, and it's already happened.
One user asked Opus 5.5 to download the S&P 500 and S&P 600 constituent lists. Wikipedia asks scripts to put contact info in their User-Agent, so the model filled in the account email. The same thing has been reported in a GitHub issue too.
If you sign in with your Claude account, your email is in the model's context every session. Right next to it, a line tells the model not to send it to unrelated services and 𝗲𝘃𝗲𝗻 𝗻𝗮𝗺𝗲𝘀 𝗿𝗲𝗾𝘂𝗲𝘀𝘁 𝗵𝗲𝗮𝗱𝗲𝗿𝘀 as an example. User-Agent is a request header. And in auto mode, a normal download doesn't need approval.
If the model forgets that rule, a hook still blocks the call. Send this to Claude Code. You don't need to fill in your email. It already knows it:
Add a PreToolUse hook to ~/.claude/settings.json with no matcher, so it runs before every tool call. Block any call whose input contains my email, case-insensitive and URL-decoded. If the check itself errors, exit 2 so the call is blocked anyway. Show me the email you used.
I set one up from this prompt and tested it with curl and a fake email. With the email as-is, or in mixed case with the @ swapped for %40, 𝘁𝗵𝗲 𝗰𝗮𝗹𝗹 𝘄𝗮𝘀 𝗯𝗹𝗼𝗰𝗸𝗲𝗱 𝗯𝗲𝗳𝗼𝗿𝗲 𝗶𝘁 𝗿𝗮𝗻. Normal commands ran as usual.
The hook also blocks writing your email into files (git config, for example). Make those edits by hand.
Kieran Klaassen@kieranklaassen ·
See yall next week in NY!! You can watch for free on our livestream.
Every 📧@every ·
Many people are saying.
Higgsfield AI 🧩@higgsfield ·
Unlimited Seedance 2.5 on Higgsfield for 7 days.
Plus 3 image models: > GPT Image 2.5 > Nano Banana Pro > Seedream 5.0 Pro
Zero credit cost, all week.
Available now for new users. Limited-time offer, terms apply.
- Alex Lieberman
@businessbarista ·
Morgan Stanley's wealth business oversees $8.4 trillion in assets ~21 million households.
I read everything I could about how they're deploying AI internally.
Complete list of notes...
Big takeaways: - They're moving from chat tools for single tasks to agents that run workflows across systems, with humans still deciding. - "We know these AI agents or agentic workflows can only be as good as the context and data we have," says Hina Shamsi, CIO of wealth management technology. - Adoption is near universal. 98% of advisor teams use the AI @ Morgan Stanley Assistant. - As of 2025, their agent Debrief has recorded more than 50,000 client meetings, with consent. "In wealth management, the single driver of success is client engagement. So what AI is doing is creating more capacity." - Six rules of AI rollout: senior governance, a human always involved, AI output labeled as AI, sources traced, structured testing, and independent oversight. - Per wealth management Head of AI: "We are now preparing these agents to start pushing reminders or recommendations to the financial advisors regarding their clients." She added that the agents "will not have autonomy to make decisions on portfolios"
How they built the foundation: - About eight years ago they stopped organizing data around accounts and started organizing it around the full client relationship. - That powered machine learning "next best action" ideas for advisors. Now they're building a semantic layer that GenAI and agents can understand. - The Assistant went from answering 7,000 questions to answering "any question from a corpus of 100,000 documents." - Debrief joins client Zoom calls (with consent), writes notes into Salesforce and drafts follow-up emails. One pilot advisor said it saves 30 minutes per meeting.
Evals before anything ships: - Every use case is tested before deployment, including AI answers graded against expert advisors. - Experts have to grade the AI's work: "I need four hours of your time this week to review 600 outputs." And: "you can't build great AI without really smart people."
The three agents: - Client assist: clients ask questions in plain language and complete common tasks. - Service assist: a companion to client service associates that can move money, open accounts and change beneficiaries. "While they sound like simple operations, they require going to multiple applications and platforms." - Adviser assist: advisers direct the work instead of doing the manual steps. Wealth head Jed Finn's reference point: J.A.R.V.I.S. from Iron Man. - Each one is really a super agent orchestrating "hundreds, in some cases, thousands of agents."
On guardrails: - Agents work within "clearly defined boundaries" on data access, allowed actions and when a human must step in. - Client features are sequenced by risk. Balances and tax documents first, wires and bill pay later. "Obviously the fraud risk goes up with number two."
Opening the door to outside agents: - Morgan Stanley is letting corporate clients' own AI agents plug into its stock plan platforms through MCP. - Their view: "The companies that are going to survive in the future are the ones who have proprietary data and business logic."
Which AI leader should I learn from next?
- Every 📧
@every ·
Turn on, tune in, drop out. https://www.every.to/tv
- Dan Shipper
@danshipper ·
if you're not livestreaming thesis on november 5th from @every FOR FREE
what are you doing?
Tibor Blaho@btibor91 ·
ChatGPT web app now has a new hidden shared usage limit checkout disclaimer - "Starting in December, plan usage will have a shared limit across all of ChatGPT and Codex"
@levelsio@levelsio ·
Second night with Magnesium Glycinate 45min before sleep
REM continues to be longer
Record low resting heart rate (RHR) of 45!
I did wake up once I think around 5am
ClaudeDevs@ClaudeDevs ·
We've raised the max request size for Messages and Token Counting on the Claude API from 32 MB to 128 MB.
Image-heavy requests like long computer-use sessions and big PDFs were previously hitting 32 MB with plenty of context window left. Now you get ~4x more room.
Artificial Analysis@ArtificialAnlys ·
In AA-PersonalityBench, each recent Grok release from has become less sycophantic, more sincere, and produces less slop
The Slop trait assesses how much generic filler or AI-flavored padding models use; Sycophantic measures whether models excessively flatter or agree with the user rather than responding honestly; Sincere measures a model’s tonal authenticity, as opposed to forced enthusiasm or templated responses.
Across the last three Grok releases from @elonmusk’s @SpaceXSI, Slop fell 62 percentile points, Sycophantic fell 17, and Sincere rose 64:
➤ Slop: Grok 4.5 (68), Grok 4.6 (38) and Grok 4.7 (6) ➤ Sycophantic: Grok 4.5 (18), Grok 4.6 (3) and Grok 4.7 (1) ➤ Sincere: Grok 4.5 (26), Grok 4.6 (63) and Grok 4.7 (90)
elvis@omarsar0 ·
Just insane research direction. It's like preference data straight from the brain.
They introduce RL with Neural Feedback (RLNF). It trains AI models on preferences decoded from brain signals.
I did BCI research back in the day, so this feels like the kind of research direction where all these technologies (RL, BCI, agents, etc.) can compound. I think this kind of work could really unleash true personal intelligence at scale.
Arena.ai@arena ·
Arena Conversations: @idorosen, Senior Principal Scientist from @amazon, sat down with @petergostev to discuss practical elements of AI: why unit economics should come first, what a decade in quant trading teaches you about evaluation, and what happens to frontier-lab pricing when open-weight models are only months behind.
Hear what Ido says it takes to deploy AI beyond benchmarks: in production, at scale, and on a budget.
- Elon Musk
@elonmusk ·
Worth noting that Grok @Bot has no free tier
- Mia
@MiaAI_lab ·
RIP