ec109685

10 ms·

121.
by ec109685 1y ago
These are poor examples: > Why do I need to know what quarks are, or how steam engines work? Why do I need to know anything about anything? I can just look it up with AI Literally all possible with Google and YouTube.
122.
by ec109685 1y ago
Yeah and it’s creative with its grepping.
123.
by ec109685 1y ago
How can you say this about a product that has 700M MAU’s: > As is true with a good many tech companies, especially the giants, in the AI age, OpenAI’s products are no longer primarily aimed at consumers but at investors
124.
by ec109685 1y ago
It greps around the code like an intern would. You have to have patience and be willing to document workflows and correct when it gets things wrong via CLAUDE.md files.
125.
by ec109685 1y ago
So much spin. Bottom line is that it’s embarrassing for GitHub to have lost the ai coding race (or to be losing badly).
126.
by ec109685 1y ago
How does Perplexity Comet and Dia not suffer from data leakage like this? They seem to completely violate the lethal trifecta principle and intermix your entire browser history, scraped web page data and LLM’s.
127.
by ec109685 1y ago
Read 5.2 They don’t add a new loss over the probe signal. Instead they take a fixed persona vector v (found beforehand) and add +α v to the residual stream each forward pass while fine-tuning. The idea is to cancel the gradient push toward
128.
by ec109685 1y ago
In almost all cases the back swipe in the spa resets you to the top of the page, navigating out of the app and back in doesn’t work, etc. It’s really hard to build a multi page spa that feels good.
129.
by ec109685 1y ago
If the key unlock for self-driving is maximizing real-world training data, then Tesla’s approach of using cameras make sense since it allows them to utilize their entire fleet to deploy incrementally better FSD to.
130.
by ec109685 1y ago
The independently wealthy part is strange, but you only live once so folks should find the path satisfies them. As for caring for a newborn, that is the least impactful moment you have with your kids. Seems like he made a reasonable trade-o
131.
by ec109685 1y ago
No releases since June 13th: https://help.openai.com/en/articles/11428266-codex-changelog It seems like Claude code via screen / multiple terminal windows is strictly better than the Codex approach given it k
132.
by ec109685 1y ago
Are PWA’s better and more popular on Android?
133.
by ec109685 1y ago
They relaxed that for browser makers: https://developer.apple.com/documentation/browserenginekit/p...
134.
by ec109685 1y ago
Lots and lots of reinforcement training.
135.
by ec109685 1y ago
They have got get their act together from a structure standpoint or these types of acquisitions are going keep failing.
136.
by ec109685 1y ago
https://x.com/garrytan/status/1941553682736439307 The thesis is that once you’re paying $200 a month, you’re beholden and won’t pay and compare it with anything else.
137.
by ec109685 1y ago
Tab completion in cursor lets you keep hitting tab and it will jump to next logical spot in file to keep editing or completing from.
138.
by ec109685 1y ago
Background of how it works: https://kirshatrov.com/posts/claude-code-internals Prompt: https://gist.github.com/transitive-bullshit/487c9cb52c75a970...
139.
by ec109685 1y ago
Good analysis. And Claude code itself will be mercilessly copied, so even if another model jumps ahead, small switching cost. That said, the creator of Claude Code jumped to Cursor so they must see a there there.
140.
by ec109685 1y ago
It seems like search access is more valuable these days since reasoning requires realtime access to site data.
141.
by ec109685 1y ago
To add to this, BeReal flamed out because it was a gimmick that was quickly copied and then forgotten by all the other platforms. Agree that blaming VCs for why TikTok has an algorithmic feed and hundreds of millions of users is simplistic.
142.
by ec109685 1y ago
Because social networks are boring without users.
143.
by ec109685 1y ago
How is this getting out of the way: “In markets like Texas, the wholesale price of electricity is set equal to the price of electricity from the most expensive generator needed to meet demand, often referred to as the marginal generator.”
144.
by ec109685 1y ago
I don’t think addiction is the right way to look at it. I think most people can easily do all the way through #6 if they put their mind to it. It’s not a physical addiction. The real cost is when you’re not intentionally trying to deprive y
145.
by ec109685 1y ago
Good point. The 23% in the example refers to the worst case where 5 tests are all null throughout the period.
146.
by ec109685 1y ago
Yes, but assuming it was enhancing something already there, it was all pointless work.
147.
by ec109685 1y ago
Keep in mind that Frequent A/B tests burn statistical “credit.” Any time you ship a winner at p = 0.05 you’ve spent 5 % of your false-positive budget. Do that five times in a quarter and the chance at least one is noise is 1 – 0.95⁵ ≈
148.
by ec109685 1y ago
In two separate private browser windows, I was identified as unique, so does that mean a fingerprint across private browser tabs would not work?
149.
by ec109685 1y ago
“Colloquially, there is an idea that autonomous vehicles are something that will erode driver jobs and put drivers at risk. And I think the irony of what we’ve seen is that it’s actually quite expensive to run an AV” This seems like a tempo
150.
by ec109685 1y ago
Because people make stupid correlations for clicks.

More ›