contrary to the idea that ai is going to replace work, i now find myself so overwhelmed with what’s possible that prioritization has become the most important part of my workflow. i could spend all day thinking about stuff to work on. every completed task spawns 10 more ideas
Marc Andreessen on why it’s critically important to expose yourself to risk:
“Put yourself in situations where you will succeed or fail by your own decisions and actions, and where that success or failure will be highly visible.
Why?
If you're going to be a high achiever, you're going to be in lots of situations where you're going to be quickly making decisions in the presence of incomplete or incorrect information, under intense time pressure, and often under intense political pressure.
You're going to screw up - frequently - and the screwups will have serious consequences, and you'll feel incredibly stupid every time.
It can't faze you.
You have to be able to just get right back up and keep on going.
That may be the most valuable skill you can ever learn. Make sure you start learning it early.”
🚨BREAKING: Microsoft Research + Salesforce just dropped a paper that should scare every AI builder.
They tested 15 top LLMs GPT-4.1, Gemini 2.5 Pro, Claude 3.7 Sonnet, o3, DeepSeek R1, Llama 4 across 200,000+ simulated conversations.
Single-turn prompt: 90% performance.
Multi-turn conversation: 65% performance.
Same model. Same task. Just... talking normally.
The culprit isn't intelligence. Aptitude only dropped 15%.
Unreliability EXPLODED by 112%.
→ LLMs answer before you finish explaining (wrong assumptions get baked in permanently)
→ They fall in love with their first wrong answer and build on it
→ They forget the middle of your conversation entirely
→ Longer responses introduce more assumptions = more errors
Even reasoning models failed. o3 and DeepSeek R1 performed just as badly.
Extra thinking tokens did nothing.
Setting temperature to 0? Still broken.
The fix right now: give your AI everything upfront in one message instead of back-and-forth.
Every benchmark you've seen was tested on single-turn prompts in perfect lab conditions.
Real conversations break every model on the market and nobody's talking about it.
NVIDIA just dropped PersonaPlex-7B 🤯
A full-duplex voice model that listens and talks at the same time.
No pauses. No turn-taking. Real conversation.
100% open source. Free.
Voice AI just leveled up.
https://t.co/YfzFQfBzMS
Everyone's reading "developers haven't written code since December."
Here's what I'm reading:
Design doc → AI builds → QR code preview → engineer verifies → merge to prod.
That's not "no code." That's a full verification pipeline with AI doing the typing.
Spotify didn't remove developers. They promoted them to architects. The question every engineering leader should be asking right now: do you have the orchestration layer to do this? Or are your devs still prompt-and-praying in a chat window?
This is the best explanation I have heard of how AI is impacting the software landscape. Not just the stocks, but the actual fundamentals of the businesses underneath
From AI Czar David Sacks himself
"You take a product like Salesforce that deals with all your customer contracts and revenue. You are not going to replace that with code that has been spit out of a coding assistant that has not been fully vetted
Think about how many bug reports on Salesforce's code base over the last 25 years. Maybe millions of them. That system has been tested across thousands of large customers and enterprises
The idea that you are just going to rip out that system and replace with code that has been probabilistically generated by an AI engine yesterday, with a small team to maintain it internally, just does not seem realistic to me"
@bgurley@rohanpaul_ai Market share, which is important for raising VC money, which is what is subsidizing inference costs currently.
The dynamic of the market has kept prices artificially low