I maintain a fork of sqlite-vec (because there hasn't been activity on the main repo for more than a year): sqlite-vec is great for smaller dimensionality or smaller cardinality datasets, but know that it's brute-force, and query latency scales exactly linearly. You only avoid full table scans if you add filterable columns to your vec0 table and include them in your WHERE clause. There's no probabilistic lookup algorithm in sqlite-vec.
You're absolutely right—sqlite-vec currently only supports brute-force search, and its latency does scale linearly with dataset size. We did some rough comparisons using its benchmark tools: on the SIFT dataset, latency was around 100ms; on GIST, it was closer to 1000ms. In contrast, with zvec's HNSW implementation, we get ~1ms latency on SIFT and ~3ms on GIST, while achieving recall@100 of 99.9% on SIFT and 97.7% on GIST.
You're right that we didn't include sqlite-vec in our initial benchmarks—apples-to-apples comparisons are always better. I've actually added basic zvec tests to my fork of sqlite-vec (https://github.com/luoxiaojian/sqlite-vec), so feel free to give it a try. We'll also be publishing a more complete performance comparison in an upcoming blog post—stay tuned!
You joke but you’d be amazed at the Reddit front page. It’s hard to tell anymore if the comments are even people, but I have noticed many fake posts of some Trump tweet he never actually made getting traction.
It’s so easy to verify his public statements. Did he really say that? Just go look.
Yet time and time again people get baited into rage mode. It’s more satisfying to post than it is to do 30 seconds of research.
> ChatGPT had 800m MAU at one report, but that's a chat interface and free. Do you really believe over half of those users are going to convert from "free" to paying $60/mo for access to the chat
Even if these things worked great for everyone, the percent of free uses who convert to paid users is low single digits per cent. For OpenAI to have any chance of breaking even in the consumer space, they need to develop an ad biz that makes around 20-25% of G does. That's a tall order in that G doesn't make good dough from search anymore as SERP page clicks are down 80% with AI summaries being good enough for most.
And let's not forget that for the bubble to sustain itself, people would currently use different LLMs would need to create a separate account in each one. There's absolutely no way most people will be paying more than one LLM unless they have a lot of disposable income.
Consumer spending is strong and growing, don't listen to dregs milking upvotes on the internet, people will easily come up with 4-5 hours of minimum wage pay in a month to cover the cost of the thing they use many times a day.
I don't use AI for anything in my private life, only at work. And I can't really imagine what it could do for me. In no scenario am I paying a monthly subscription for it.
More importantly, can Clawdbot even reliably access these sites? The last time I tried to build a hotel price scraper, the scraping was easy. Getting the page to load (and get around bot detection) was hard.
That’s why the author explains that the page loads in a real Google Chrome instance on a real Mac mini from the same residential IP as his other devices.
You do know that bot blockers keep track of metrics besides user agent and IP address? Hotels and concert ticket selling websites use some of the most aggressive bot blockers out there.
I do. Most of these bot blockers block bots because of scale: these bots operate with superhuman speed and their traffic comes from all sorts of IP addresses. Tools like Anubis appear because such bot traffic dwarfs human traffic. And they typically have fake User-Agent headers set to a browser while their TLS/HTTP fingerprints would suggest they are made from curl or the requests library.
This is different. There is no scale. The bot’s browsing session exactly replaces a human session. The browser is real.
Well I read the article. It seemed to me that the way the author is using OpenClaw is trivially done manually. The author just didn’t want to use the computer and preferred to chat with an AI. You might think there is no point and I would agree.
it uses a real chrome browser window (that i can see when i remote desktop into the mac mini) that's been very good so far.
re: scale, is it that this whole project is worthless bc nobody needs it, or is it that its so good that scale is a requirement? this is a project i built for myself, i'm not commercializing anything
> When the sheriff's department looked into the case, they took the opposite actions. They charged two of the boys who'd been accused of sharing explicit images — and not the girl.
reply