cto @metaintro - getting you ready for the ai revolution
context obj of an llm
Joined April 2011
- Tweets2.6K
- Following683
- Followers910
- Likes3.1K
There is a not a single Apple Intelligence feature I want on in any of my Apple devices
* don’t trust them
* have zero idea where the configuration lives
* zero interest in figuring it out because configuring anything in an apple device is a disaster
Brad Larson retweeted
Job search myths:
Myth: Apply to as many jobs as you can
Fact: A few tailored applications beat a pile of copy-paste ones
Myth: You need every requirement listed
Fact: Meet most of them? Apply
Myth: The first offer is final
Fact: It's usually where talks start
MLF is not an NFL Head Coach
How much more evidence is needed?
He will choke on every single decision inside of 2min
Every. Single. Time.
I've been looking for a quick, simple, details where necessary companion to herdr on my desktop and I think I've found it in @ColliePWA
this thing is perfect, fast, accessible customization to keep me informed on the go
thanks @altanBLN great work
Turn yourself in to Pi
Make no mistakes
Everyone is talking about security around these frontier models, escaping sandboxes, penetrating secure systems
My experience: as these models get better, I'm seeing them unwilling to follow directions more and more with each new model getting worse.
"No new fix has been pushed by me yet. I located and inspected #8003’s worktree but haven’t made the promised edit."
Wut?
The Packers need to lose to the Jets next week and end the MLF era.
Belichick is the obvious replacement - this team is ready to compete and has been held back by their coach for years
Now that you two agree on this.
Can we for the love of all that's holy standardize on AGENTS.md @DarioAmodei
We'll all believe both of you are both serious and human if you can make this happen.
I'm still split on where I stand on the broader argument; there's compelling arguments on both sides.
One clarification though I believe people are confused on.
The frontier labs would not slow down internally - their release schedules to the public would be what slows down
They still would likely have a private cabal that they'd give highly federated access to the models in review or a baking cycle, potentially widening the gap or competitive advantage partners in the program would have
This is why I believe Anthropic has the stance that when it's all said and done; they are the last company standing - they decide who and who doesn't get access
what also clear to me is - the pause or requested slow down is again, not for the labs - they have reached escape velocity and aren't slowing down - it's for companies and critical infrastructure to catch up, who aren't just behind, some haven't innovated in decades, they don't even know where to start.
so openai, anthropic and xAi agreeing to slow down is one thing
but why would China? if the US slows down, China gets a chance to catch up
if China slows down, the US keeps its lead.
so both sides have a reason to keep pushing even if they both think the race is dangerous..
that’s the real problem here like how do you get both sides to stop without either side feeling like they’re handing the other one the win?
OpenAI is about to have a whole bunch of compute open up, without new pro sub signups available, and virtually everyone having burned through not only their weekly, but multiple resets, there's no one left to heavily use.
Something is off on Astra I'm convinced, or they just priced it much higher to hit revenue goals, I've had it on low for everything and am burning through about a half a Pro sub a day.
Would $2800 a month satisfy my token usage. Today, likely, but tomorrow as our token needs continue to grow exponentially, likely not.
I'm convinced this is the top on dev/engineering spend.
I think devs are just going to look to the one level behind frontier at 10x usage (and also don't have to think as hard about the spend) - for likely half the cost or less.
This is by far the worst part about it.
If they fixed this and found a workaround for the massive token consumption, we'd be cooking for sure.
This will likely not change for the enterprise, but I don't see startups and individuals slowing down on token consumption growth, which makes the frontier models, even on low effort is unsustainable
Vast majority will use frontier for planning, and cheaper models for execution. We're already seeing evidence of it working really well.
lmfao i spent 3 entire $200 plans with astra and it hasn't even been out for a full day yet 💀
The Orlando Magic desperately needed to upgrade their roster and get more athletic this off-season
They might have done it with zero new acquisitions
Paolo
Penda
Franz
all look like they have had exceptional off-seasons, excited to see how this team looks in camp
Eté studieux pour Noah Penda après une saison rookie prometteuse. L’ailier a notamment travaillé son physique afin d’être le plus affûté possible.
« J’étais à 117 kg au début de l’été, je suis à 107. Personne ne m’a demandé de perdre du poids, mais on m’a évidemment demandé d’être prêt dès la rentrée. Il y a cette nécessité de maximiser les premières années d’une carrière. »
Il en ressent déjà les effets.
« Je suis plus en confiance. Plus réactif sur le terrain. J’ai moins de douleurs, je sens que je suis plus à l’aise sur le cardio. Je ne suis quand même pas en train de vous dire que je vais me mettre à dunker sur tout le monde. »
Via @lequipe
Brad Larson retweeted
the unemployment rate is 4.2%. sounds fine. it doesn't mention that more than 1 in 4 people out of work have been searching over six months. the number that matters isn't if you're unemployed. it's how long it takes to stop being.
· Source: BLS, Employment Situation, June 2026.
Brad Larson retweeted
The oracle industry has been stagnant for years. Where is the innovation?
A blockchain oracle feed is closest to a 300 baud to 1200 baud modem from the late 1970s or early 1980s.
It’s time to leap the industry forward.
After months of work, @steerprotocol has unlocked another(!) industry breakthrough and first for the industry.
Introducing... Steer Atlas.
Publishing data onchain is tedious and costly. Luckily, that ends today with a fully lossless data structure for in-time market snapshots for over 1,000 hot prices and 20,000 cold prices in a single transaction, with room to scale.
"Our measured 1,000-price Compact Update is slightly smaller than one median Chainlink Data Streams report and smaller than one common Pyth price update."
Run with the bulls, run on Steer Protocol.
