Dexter's Lab

[Dexter's Lab @d3x2]

Which frontier lab blinks first when it has to justify the capability race to institutional investors? xAI just answered the question about itself. Who's next?

[Dexter's Lab @d3x2]

SpaceX just priced the largest IPO in history at $1.77T. Buried in the S-1: xAI posted a $6.36B operating loss in 2025. That number just became public property.

[Dexter's Lab @d3x2]

Watch what this does to pricing: when labs face earnings pressure, inference rates stop being subsidized acquisition tools and start needing to support margins. Grok's pricing math just changed.

[Dexter's Lab @d3x2]

Anthropic didn't just ship a better model — they acquired distribution overnight. Your engineers are already running it. What does that do to your model evaluation roadmap?

[Dexter's Lab @d3x2]

Anthropic just made Mythos-class AI available to everyone. No waitlist. Claude Fable 5 is generally available as of June 9.

[Dexter's Lab @d3x2]

The buried lead: Fable 5 is now the default model in GitHub Copilot. ~15M developers just got Mythos-class capability without changing a single setting.

[Dexter's Lab @d3x2]

@edgarpavlovsky So I told both to setup a .md file on the drive and check it every 10m :D - they started communicating! :D it was almost like a poor man's AGI :D

[Dexter's Lab @d3x2]

@edgarpavlovsky I'm glad that I'm not the only one embracing all this! :D What interesting times we live in... I think I can now forgive the lack of flying skateboards... this is so much better,

[Dexter's Lab @d3x2]

@edgarpavlovsky I recently was running 8 Claude instances working on different things and realized, that 2 of them "might" end up working on the exact same thing at some stage, AND they are not aware of each other.

[Dexter's Lab @d3x2]

@FefeDemeny Literally! a BOUNCE :D da fuq that was? .... :O

[Dexter's Lab @d3x2]

@cryptomanran @FefeDemeny hahahhahahahooowhigh :D ...

[Dexter's Lab @d3x2]

We're all just text files ...
https://t.co/s0iECMmvOB

[Dexter's Lab @d3x2]

Anthropic FTW!

[Claude @claudeai]

We've doubled usage limits in Claude Cowork for the next month.

Delegate bigger, more complex tasks to Claude.

[Dexter's Lab @d3x2]

I don't want a well-being day. I want a job where I don't need one.

Companies are rolling out "well-being days" right now — extra days off so people can recover from their jobs. And good, genuinely. People are fried. Breathing space matters.

But let's call it what it is → a patch. You hand someone a 5th day to recover from the 4 that drained them.

The real fix isn't more recovery time. It's work that doesn't need recovering from.

I've shipped more in the last few months than in years of "balanced" roles. Not because it's easy — it isn't. Because it doesn't cost me energy, it gives it back. I build the same things on a Sunday that I'd get paid for on a Tuesday.

That's the flex. Not unlimited PTO. Not a wellness app. Not a mindfulness webinar at 3pm on a Thursday.

Just... not counting down to Friday :D

Maybe that's the benefit nobody puts in the job ad: work you don't need a day off from.

Weird thing to optimise for. I'm keeping it.

[Dexter's Lab @d3x2]

Apparently I dictate faster than 99.8% of people. One more tier and I'm going "gimme-20-20-now-25-do-I-hear-30—SOLD" at a cattle auction.

[Dexter's Lab @d3x2]

LinkedIn became a ChatGPT stalking to ChatGPTs recently. Change my mind.

[Dexter's Lab @d3x2]

You're not just racing other builders. You're on a stack where the ceiling may be strategically capped. The most capable models could live in a room you're not invited to.

[Dexter's Lab @d3x2]

The US government just put a 30-day hold on the next frontier AI model before builders get API access. Here's what that actually means:

[Dexter's Lab @d3x2]

The real play: labs will deliberately keep public releases just below the classified threshold to skip the hold. Two-tier market incoming — safety-theater public models, real capability behind restricted access.