slake
slake on Hacker News
Funnily I've found advanced models to be bad at calculating the length of a string.
Heathrow has especially gotten horrible. It used to be my favourite transit airport, now it's packed to the brim with barely any place to sit. The restrooms are ill-maintained.
The security opens into a narrow space with a 100 people trying to put on their belts and shoes while acres of real estate is wasted on empty Hermes, Gucci stores.
While I find British people generally pleasant to talk to, the ones at Heathrow feel like they're jail wardens.
The VCs footing the bill is really your pension funds and 401Ks and banks passing through the VCs. If VCs lose money the contagion spreads through the economy.
There goes Enterprise usage
Also if this is on the path of anything the Chinese do in the physical goods world, inference will be rockbottom cheap in a few years because they'll invest in the hell out of energy, GPUs, research, etc. The same thing they did with EVs.
Only artificial barriers will keep people using some of the frontier stuff in a couple of years. No costs will justify.
Posted by
slake on Thursday April 30, 2026 @10:32PM
from the ask-hn dept.
Just an observation. When asked to summarize articles, or extract insights, I see the word 'prior' being used a lot more by Claude than usual English language writing (Journalistic essence). And it's clearly using it in a Bayesian sense, because it's always mentioning things like 'Updating priors', 'the prior doesn't hold', etc.Probably something I noticed after reading the 'goblin' and...
I actually like 2006 twitter. Almost pleasant because of its obscurity.
- Only 140 characters - Increasing that added to the slop. Brevity was soul
- No 'For you' algo. You had to find and follow the people you wanted to follow and then read all of their posts
- RTing it was literally adding RT in front of the other person's tweet.
- There was a 'like' icon (I think - don't remember)
- There was no paid 'verified' accounts. You had to signal your official twitter account through different means.
- There was no monetization for random clickbait ragebait content.
- Ads are not that intrusive on twitter. That as a monetization strategy still works.
- You only discovered people and not content.
Bro. I'm obsessed with this. I spent a good 2-2.5 hours on a trot with this. This was one of those things I always wanted to learn a lot about and this was just perfect for that. It wasn't that easy, I really struggled and gave up on some levels but came back and struggled through them. The HexRacer type levels are difficult and I wish I could skip through those.
Also when you give a solution, you probably need an explanation of why that works.
Can't wait till you add the additional coming soon levels.
Your feedback feature seems broken. I tried giving feedback about some basic bugs and it kept saying can't send message.
Please also add a waitlist to notify when you release the additional levels.
I'm wondering if this can use different workspace folders for different projects? I'd like to manage all my side projects through this but I can't put them all in one folder.
Funnily it seems you were already more organized than I am today before you had the *Claw. That's why you were able to transition to it as a 'life operating system'.
I have too much stuff in my head and nothing written down and that's beginning to be a hindrance to the transition.
Phew. What is your system prompt? Are you 4b or 8b?
Conspiracy theory here, but all the big AI companies have to do is find some 0-days and cause some data leaks at the top SoR companies for this confidence to waver.
That's amazing. Feels like a major step ahead. No lag, very snappy. Outstanding work.
Feels like those sci-fi shows where you can talk to Hari Seldon even though he lived like a 100 years ago.
My prediction, this will become really, really big.
Wouldn't this expose your github token to anyone in the frontend?
May I ask what kind of tasks these are specifically? Is it mostly data entry, management and reporting? Or are you interfacing with devices?
Sometimes I think we need to expose what 'memories' or their semantic representation has been stored in a 'memory store' so that humans can review and verify it over time. This will help the LLM 'forget' things that the humans using it don't really think is that relevant.
This is how I don't get when some predict that we'll have a shorter working week when AI improves productivity. If productivity gains are 50% the companies would rather have 50% lesser employees rather than 50% shorter work weeks.
This is just MSFT (as did others) acknowledging the future has a lot less workers in it.
How do you do the REPL integration into the manual workflow? Could you explain that a bit?
The article doesn't answer its own question. The inherent contradictions in the ratings. Romania who just 'cancelled an election' and banned the winning candidate from contesting is 40 positions above India.
Win an election of 800M people by using tweets and shares. How delusional can one person be.
What's the database that dbt is connected to in this scenario?
Does it work with o1 type reasoning models. I had trouble running phidata (the old named framework) with it.
Even Indian software firms have begun to hire in Vietnam since they work 1.3x as hard for half the pay that an Indian does.
GCP is the worst at this in my opinion. Their UI is so 2000 and hard to use with multiple menu levels and not even good search.
I'm wondering, If I bought an emergency jumpstart battery for my car. Can I run on a failing battery for a lot longer if I just jumpstart it every time I start the car?
Usually the highest current draw is for the starting right. Post that it runs just on engine generated power?
Me too. Please accept me into your cult.
This is smart. If you're contributing to opensource Github, there's a high likeli hood it gets into one of these datasets and maybe training data for an LLM.