sroussey

7 ms·

121.
by sroussey 2mo ago
I have Opus grade each of my repos, and each subsystem of them, and do it every weekend. Since the version of opus has changed over time, so have the evaluations. I was going to constrain it, but since they edit the same files, I can alter
122.
by sroussey 2mo ago
It need only be one of several tools in a toolbox.
123.
by sroussey 2mo ago
The parent commenter was talking about Mistral as a single company and you switched from that to all of the EU. There definitely have been Chinese companies with models that fell behind, which is the more direct comparison. As for the EU in
124.
by sroussey 2mo ago
Yes, but to walk in to the market late, and not have all that debt, it will be too much of an opportunity for an aspiring company. Google was not the first search engine. But in a way, it was the last.
125.
by sroussey 2mo ago
But the incentive to produce that intelligence is so high, that many opportunities become practical to explore. And many of them show doing AI inference workloads at 1000x cheaper and with 1000x less power, and sometimes 1000x faster. If an
126.
by sroussey 3mo ago
yes, i went back and re-read the report. an immediate read 1 instruction later works, but 10 does not. the page marked dirty but pointing to the zero page. some context flapping i assume
127.
by sroussey 3mo ago
The way I read it was that it wrote to a page and then did a read on that page within 10 instructions and that was not enough time for the memory system to have gone through the process of creating the page backed with real memory and that
128.
by sroussey 3mo ago
And this explains a lot of politics these days…
129.
by sroussey 3mo ago
1970s design.
130.
by sroussey 3mo ago
Yes, there is always a failure mode. But if do a search or have some result set, it is convenient to have paging work as close to expected as possible. Say you are processing/shipping orders. You do the first page and then advance to p
131.
by sroussey 3mo ago
That does not work for web apps, mobile, etc
132.
by sroussey 3mo ago
Even then, that does not work if new records are being inserted.
133.
by sroussey 3mo ago
Google already has a system where data and access control are combined as part of shared docs. I think Zanzibar for docs themselves on GDrive. Not sure the name of the system that manages a portion of a document in a shared doc which coloca
134.
by sroussey 3mo ago
The shots are expensive, but the pills are free in the USA. They do have adverse kidney issues over prolonged use (more for some people than others).
135.
by sroussey 3mo ago
The money being poured into AI infrastructure means there is a market for new ways of doing things that take 1/1000 of the power or are 1000x faster, or both. And there are many such moonshot startups. AI on GPUs is an efficient as gam
136.
by sroussey 3mo ago
With the exports rewritten? Maybe we should have a export of “ts” like we have “types” instead?
137.
by sroussey 3mo ago
Yeah, 1.3.14 has some bad regressions. 1.3.11 is ideal for tests CI. Or canary. For production, I don’t have any advice. There will be no 1.3.15.
138.
by sroussey 3mo ago
There is always AssemblyScript
139.
by sroussey 3mo ago
Publishing a lib as JS with .d.ts type files gives you JS compatibility and TS types for using the lib, but no TS source so no types inside. I’ve seen people want to publish TS libs but there are issues: TS will type check the libs as well
140.
by sroussey 3mo ago
In that case, just assume linear scaling? You only need pay attention to a single core.
141.
by sroussey 3mo ago
It took us a bit of work back in the day to use Firebug to debug Firebug. But pretty easy these days to have Chrome dev tools to debug itself.
142.
by sroussey 3mo ago
Look at Meta’s capital budget and tell me people aren’t using AI. People use Facebook, Instagram, etc. I’d say the same for Google, but with Google Cloud it’s hard to know where the dollars are going and for whom. Who exactly is the 2% figu
143.
by sroussey 3mo ago
Then subsidize food directly rather than via water.
144.
by sroussey 3mo ago
My stuff is on Vercel, so it’s more of an issue as it uses up my limits. I’m moving to Cloudflare and I want to see how changes I make affect their bot, if at all.
145.
by sroussey 3mo ago
Everyone using Google. Watching Netflix. Listening to Spotify. Shopping at Amazon or a Shopify store. Or reading almost anything these days. :( Advertising they see. Using a credit card, writing a check. Beam forming wireless signals. It’s
146.
by sroussey 3mo ago
Water usage for a data center is nowhere near any agricultural use (which is subsidized in California so they aren’t incentivized to change much). Data centers don’t need water at all. It’s just the municipality offering cheaper way to cool
147.
by sroussey 3mo ago
The data centers in L.A. are in old bank buildings. I used to locate my servers at One Wilshire. Water is too expensive and wasteful to just run through pipes and send out to the ocean just for cooling. Power though… the reason bank buildin
148.
by sroussey 3mo ago
> 2% for one thing that most people hardly use at all AI is hardly used at all? Basically every company and consumer in the country is using it, transitively if not directly.
149.
by sroussey 3mo ago
Yes, indeed! I have a site where 99% of the traffic is Claude. I could ban them but I don’t bother.
150.
by sroussey 3mo ago
Or… war with China is inevitable if only to keep the gap in military capabilities. There is a thinking to mow down their military while we still can in some circles. Like those in power currently. Scary.

More ›