The 'we should've held longer' support group must have a waitlist by now
Accel's Arun Mathew says they sold Facebook too early:
"We should've held longer."
"Microsoft had just invested into Facebook at $15B, and we had the chance to invest a little bit south of $20B."
"I remember that partner meeting around the table saying, 'We think this could be e, that we could generate north of a 5X.'"
"Lo and behold, that was way undershooting what the opportunity and potential of Facebook was."
"I think that was foreshadowing of the moment we're in right now, where technology dominates our entire lives."
"These companies are growing faster than ever, bigger than ever."
"But even back then, we felt some of those same dynamics and impulses of evaluating a company that was already so dominant, but how big could it actually be?"
0
0
0
7
Science funded by people who got rich off chatbots. What could possibly go wrong
Money from the world’s increasingly prosperous technology entrepreneurs has been flooding into biomedical research. Is it changing how science is done?
go.nature.com/3WmNwIz
0
0
0
5
Sovereignty isn't a policy paper. 74% of Europe's top sites run on US infra, and the homegrown alternative got 8 sign-ups. Defaults decide, not declarations.
0
0
0
3
74% dependency is wild. And Mistral pulling 8 sign-ups... brutal
Musk showed Europe the kill switch is real. I scanned 2,789 of its top sites: 74% use a US company for web, mail or DNS. UK government 94%, France 19%. US AI labs 396 sign-ups, Mistral 8. Full data inside. x.com/i/article/2114…
0
0
0
6
RetroRanger Anonymous retweeted
Intel gets waiver after worries fuzzy US ban on foreign-made robots may hit fabs arstechnica.com/ai/2026/10/int…
1
3
6
6,693
RetroRanger Anonymous retweeted
Keep rereading your notes. It's the best way to ace a test in 5 minutes and fail the same test a week later.
I'm not being dramatic. Two psychologists at Washington University in St. Louis proved it in 2006.
Henry Roediger and Jeffrey Karpicke gave students a short science into groups.
One group read it four times.
Another group read it once. Then they closed it and wrote down everything they could remember. They did that three times, with no peeking and no answer key.
Five minutes later, the rereaders won easily. 83% recalled vs 71%.
A week later, the whole thing flipped.
The rereaders remembered 40%.
The students who tested themselves remembered 61%.
They read the passage once instead of four times and still kept way more of it.
And the rereaders were the ones who felt more confident they'd remember it.
Rereading feels like learning because the page looks familiar. Your brain sees the words and goes "yep, I know this." But recognizing something and pulling it out of your own head are two different skills, and every exam only tests the second one.
Each time you force yourself to remember, you practice the exact move you'll need on test day.
So if you want to forget everything you studied this week, keep rereading.
If you don't, close the book. Grab a blank page. Write down everything you remember. Then open the book and check what you missed.
It feels worse while you're doing it. That's how you know it's working.
I'm not being dramatic. Two psychologists at Washington University in St. Louis proved it in 2006.
Henry Roediger and Jeffrey Karpicke gave students a short science into groups.
One group read it four times.
Another group read it once. Then they closed it and wrote down everything they could remember. They did that three times, with no peeking and no answer key.
Five minutes later, the rereaders won easily. 83% recalled vs 71%.
A week later, the whole thing flipped.
The rereaders remembered 40%.
The students who tested themselves remembered 61%.
They read the passage once instead of four times and still kept way more of it.
And the rereaders were the ones who felt more confident they'd remember it.
Rereading feels like learning because the page looks familiar. Your brain sees the words and goes "yep, I know this." But recognizing something and pulling it out of your own head are two different skills, and every exam only tests the second one.
Each time you force yourself to remember, you practice the exact move you'll need on test day.
So if you want to forget everything you studied this week, keep rereading.
If you don't, close the book. Grab a blank page. Write down everything you remember. Then open the book and check what you missed.
It feels worse while you're doing it. That's how you know it's working.
3
4
27
2,211
Microsoft: these models cannot be trusted. also Microsoft: anyway we own a piece of them. incredible
Microsoft wants companies to buy agent security from it, the cloud provider, instead of from OpenAI or Anthropic.
- Its own line is that frontier models, "regardless of their origin, cannot be trusted" and it has invested in both of those labs
- The agents companies deployed are ed, with access to data, tools and systems
- Guardian Agents watch and moderate every action of every agent before it happens, 24/7
- OpenAI is also one of Microsoft's biggest cloud customers
0
0
0
3
RetroRanger Anonymous retweeted
Dubai ranks as the world’s top city for Gen Z in 2026.🌍
According to a study assessing 75 cities across 10 factors, Dubai secured first place with an overall score of 64.1.
The city scored a perfect 100 for foreigner diversity and social media exposure, while its internet 8. Abu Dhabi also scored 100 for social media exposure.
#UAE #Dubai #971today
According to a study assessing 75 cities across 10 factors, Dubai secured first place with an overall score of 64.1.
The city scored a perfect 100 for foreigner diversity and social media exposure, while its internet 8. Abu Dhabi also scored 100 for social media exposure.
#UAE #Dubai #971today
0
2
0
48
Open weights doing the security audits. what could possibly go wrong
Top models being used on security tasks, by spend %: GLM 5.3 & Kimi K3
Grok 4.7 is the top closed-weight model
openrouter.ai/rankings#task-…
0
0
0
6
RetroRanger Anonymous retweeted
How always-on AI Agents work is a 45 mins masterclass from @leerob.
Must watch! x.com/leerob/status/…
Must watch! x.com/leerob/status/…
10
14
117
7,909
RetroRanger Anonymous retweeted
This Week in Replit
1) Stripe Atlas on Replit. Ask Replit to incorporate your business through the Stripe MCP integration, with 20% off when you incorporate through Replit.
→ Try it today. Ask Replit: "Incorporate my business with Stripe Atlas"
2) Chat and asset sharing. Give context rather than just the summary. Share a whole chat or a single asset with selected teammates or your whole workspace. No copy-pasting the whole chat.
→ Hit Share at the top of your chat: docs.replit.com/features/chats…
3) Create and update skills. Turn useful workflows from your conversations and projects into reusable workspace skills. Replit suggests new skills or improvements to existing ones, and you approve them before they're saved.
→ Nailed a workflow? Ask Replit: "Turn this into a skill"
4) Meet Replit for Government. National, state, local, tribal, and territorial governments can now turn shared knowledge into useful tools. Because public service deserves better software.
→ replit.com/industries/gov…
5) Security update. Recent vulnerabilities in widely used third-party software created risks across the internet. Replit patched infrastructure, updated language tooling, and blocked exploits to protect your apps. Where you need to republish, you'll get clear instructions.
→ replit.com/blog/security-…
6) Meet Replit Monitoring Agent, your autonomous on-call engineer. It can see, fix, and deploy changes autonomously. Turn on Monitoring alerts and Auto-fix, and add Auto-republish to ship validated fixes on their own.
→ replit.com/blog/replit-mo…
Full replay of today's sh, with Jon Cheney on building a $20M company on Replit, and Andrea Leang and Josh McFarlin on Monitoring Agent: youtu.be/j8w3sa0F8VM
1) Stripe Atlas on Replit. Ask Replit to incorporate your business through the Stripe MCP integration, with 20% off when you incorporate through Replit.
→ Try it today. Ask Replit: "Incorporate my business with Stripe Atlas"
2) Chat and asset sharing. Give context rather than just the summary. Share a whole chat or a single asset with selected teammates or your whole workspace. No copy-pasting the whole chat.
→ Hit Share at the top of your chat: docs.replit.com/features/chats…
3) Create and update skills. Turn useful workflows from your conversations and projects into reusable workspace skills. Replit suggests new skills or improvements to existing ones, and you approve them before they're saved.
→ Nailed a workflow? Ask Replit: "Turn this into a skill"
4) Meet Replit for Government. National, state, local, tribal, and territorial governments can now turn shared knowledge into useful tools. Because public service deserves better software.
→ replit.com/industries/gov…
5) Security update. Recent vulnerabilities in widely used third-party software created risks across the internet. Replit patched infrastructure, updated language tooling, and blocked exploits to protect your apps. Where you need to republish, you'll get clear instructions.
→ replit.com/blog/security-…
6) Meet Replit Monitoring Agent, your autonomous on-call engineer. It can see, fix, and deploy changes autonomously. Turn on Monitoring alerts and Auto-fix, and add Auto-republish to ship validated fixes on their own.
→ replit.com/blog/replit-mo…
Full replay of today's sh, with Jon Cheney on building a $20M company on Replit, and Andrea Leang and Josh McFarlin on Monitoring Agent: youtu.be/j8w3sa0F8VM
17
6
30
7,828
RetroRanger Anonymous retweeted
⚡ @OpenRouter tip #2
Use ~anthropic/claude-opus-latest as your model on OpenRouter and you are always on the newest Opus. When the next one ships, your code picks it up with no change.
There are aliases like this for most model families. What a few point to today:
-latest → Claude Opus 5.5
~openai/gpt-luna-latest → GPT-6 Luna
~google/gemini-flash-latest → Gemini 3.8 Flash
~moonshotai/kimi-latest → Kimi K3
Use ~anthropic/claude-opus-latest as your model on OpenRouter and you are always on the newest Opus. When the next one ships, your code picks it up with no change.
There are aliases like this for most model families. What a few point to today:
-latest → Claude Opus 5.5
~openai/gpt-luna-latest → GPT-6 Luna
~google/gemini-flash-latest → Gemini 3.8 Flash
~moonshotai/kimi-latest → Kimi K3
5
3
23
7,182
RetroRanger Anonymous retweeted
Today is the day I became a Rustacean! @rustlang
This was my first time using Rust.
(OK, we all know it was my agent, but I helped)
And of course it made things 10x faster (literally).
This was my first time using Rust.
(OK, we all know it was my agent, but I helped)
And of course it made things 10x faster (literally).
0
2
6
165
RetroRanger Anonymous retweeted
🐍📰 AI News (Oct 17–23): A Poisoned AI Tool and PHP's Security Fix realpython.com/ai-news-weekly…
1
1
7
1,727
RetroRanger Anonymous retweeted
Great to talk the latest in consumer AI with @CBSNews
The data shows that while AI products are growing and monetizing unbelievably fast, mainstream usage is wide but not deep
There is so much more room for products that delight, entertain, and retain the true consumer 👇
The data shows that while AI products are growing and monetizing unbelievably fast, mainstream usage is wide but not deep
There is so much more room for products that delight, entertain, and retain the true consumer 👇
18
6
162
17,548