pbadams

There's the Khadas Mind series of mini pcs. They have a proprietary docking interface though. Agree that it would be great if this form-factor was more common.

The weight and reinforcement of trees, poles, deer, etc. are difficult to change. Single-vehicle crashes cause more than half the crash deaths in the US[1].

I don't think the safety record of a motorcycle is a standard to aim for -- but even then a kei truck is at least twice the weight and likely has significantly less sophisticated brakes.

[1] https://www.iihs.org/topics/fatality-statistics/detail/state...

It's not mentioned in the original paper, but DiskANN also supports PQ at build-time via `--build_PQ_bytes`, though it's a tradeoff with the graph quality as you mention.

One interesting property in benchmarking is that the distance comparison implementations for full-dim vectors can often be more efficient than those for PQ-compressed vectors (straight-line SIMD execution vs table lookups), so on some systems cluster-and-merge is relatively competitive in terms of build performance.

I think you might be right that encouraging curious discussion of these topics is important to the mission of HN.

However, if the moderation approach is going to change, I think it would be better to do so explicitly through changes to the site guidelines rather than in an ad-hoc way. I don't think that this article is covering an 'interesting _new_ phenomenon' (emph. mine) as discussed in the guidelines, and indeed most of the comments are talking about moderation policy or the conflict in general as opposed to the details presented in the article. Perhaps it would be better to have a thread explicitly focused on members of the community engaging with each other as individuals, such as a hypothetical 'Ask HN: How has conflict personally affected you?' or 'Ask HN: How/why have your views on this conflict changed over time?'

The stories that the article has to tell are important, but they aren't the thing that people are discussing here. And moderating submissions instead of explicitly discussion-focused posts invites some of the concerns about sourcing and bias that have been raised in other comments.

Something I don't fully understand, from [1], Altman was an employee of the for-profit entity. So to fire him, wouldn't the non-profit board be acting in it's capacity as a director of the for-profit entity (and thus have a fiduciary duty to all shareholders of the for-profit entity)? Non-profit governance is traditionally lax, but would the other shareholders have a case against the members of the non-profit board for acting recklessly w/ respect to shareholder interests in their capacity as directors of the for-profit?

This corporate structure is so convoluted that it's difficult to figure out what the actual powers/obligations of the individual agents involved are.

[1] https://openai.com/our-structure

Altman has claimed before that he doesn't hold equity in OpenAI. He could have some kind of more opaque arrangement that gives him a material stake in the financial success of OpenAI, and downplayed it or didn't disclose it to the board.

Who knows, though -- I'm sure we'll find out more in the next few weeks, but it's fun to guess.

There are a lot of directions people try to go, making different tradeoffs in the complexity of the clustering, the loss from the quantization, the impact on performance (esp. trying to get some subset of the tables to fit in cache). Readers might be interested in [1], which gives a survey of some of the directions.

In general though PQ is a pretty good baseline. I'm glad all these vector DB companies seem to have decided that the best form of marketing is high-quality summaries/tutorials about fundamental concepts, it's a good contribution to the community.

[1] Fig. 1 in https://www.jstage.jst.go.jp/article/mta/6/1/6_2/_pdf (2018)

Apart from the comments you've already gotten, another goal of geospatial systems is to support range queries (e.g. for the bounding box of the user's screen, what are all the businesses in that box). In higher dimensions range queries are mostly useless and the focus is on NN queries.

But as the other comments have mostly said, it's mainly dimensionality and scale differences that drive the design differences (e.g. graphs end up working better than trees in high dimensions)

> Let's put deep learning aside for a second. Can anyone here name a single academic in any other subfield of computer science who had a string of groundbreaking research works where every one of those works is from 1980 or after?

Shafi Goldwasser (Probabilistic encryption, zero-knowledge proofs, etc.) got her PhD in 1984.

Leslie Lamport is just a bit before your deadline, his 'Time, Clocks' paper came out in 1978, but the bulk of his work was after 1980, including paxos.

While there's definitely some truth to your Kuhnian view of 'times of revolution' in a field, I think it's hard to apply that to recent progress because it may just be that it's not clear which research works were groundbreaking without the benefit of hindsight. To me, the revolutionary period of CS research is still ongoing.

Part of the difficulty is, if very few people are browsing to page 2, deciding what to put on page 2 becomes harder and harder.

Google has a lot of user behavior signals to decide what should be in results 1-10. Deciding if a page should be ranked 20, 200, or 2000 without any user clicks to check if you're right is really difficult.

I would bet that since 2008/9, the relative numbers of spam site operators, Google engineers, second-page searches have changed significantly.

There are also fewer buses needed to carry the same number of people, and they are operated by professional (and hopefully less distracted!) drivers.

I'm not sure a good data source, but I'd expect that the serious injury rate per person-mile of buses is much lower than that of private cars.

Even if the 'bug' of limited time off is solved, I think people's valuation of their time is still pretty high. The trains/airships would have to be quite nice (read: heavy, expensive, possibly less efficient) for you to be willing to spend so much time in them.

On a sillier note, I wonder what the fastest speed you can reach is, in terms of timezones/hr.

My guess was a polar route like ANC-HEL, but there don't seem to be any direct passenger flights. Condor flies ANC-FRA seasonally, which is pretty much exactly 1 hour time difference per hour of flight. Another possibility is TAS->URC, at 1.2.

The author's routes, YVR-LHR and YVR-NRT, are both 0.8-0.9 hours difference per hour flight time, depending on direction.

I agree in the typical case, but they support concurrent add/delete one of their index options. Handling consistency/contention for modifying whatever graph/tree/etc structure they are using is probably nontrivial, and the resulting cache invalidations would also likely affect the QPS.

P.S. Great work on your site, by the way - it's a really inspiring project!

They used to publish some benchmarks on their site, but seem to have removed them. You can find them on archive.org[1]. I guess it is understandable, since vector search performance is pretty unpredictable, and depends on a lot of factors. If their target market is people who want vector search without needing to read a bunch of papers first, benchmarks might be more confusing than they are helpful.

edit: While I do think it's understandable, it's not great for transparency. Even if they don't want to open-source their index, I would admire it if they were willing to give ann-benchmarks[2] an API key to publish some independent results.

Disclaimer: I work on vector search at a different company

[1] https://web.archive.org/web/20210227105542/https://www.pinec... [2] https://github.com/erikbern/ann-benchmarks

Most vector search research teams are a lot smaller than you suggest, and haven't been around that long (e.g. the FAISS paper was published in 2017).

From public info, you can see they have at least one researcher working there. It's believable to me that they could have some new innovations, especially since the product space they're focusing on is different from other teams working on vector search. State-of-the-art for a specific set of constraints is still state-of-the-art.

However, considering how much of their edu-marketing content is posted to HN, it would be great if they could share more details about the internals of their index with the community. One of the great things about vector search is how many techniques are open sourced or documented in papers :).

Disclaimer: I work on vector search at a different company

In addition to these other answers, a commonly used metric in the literature is the k-Recall@R, or the portion of the top k ground-truth vectors that are found in the top R ANN queries. This parametrization can make it clearer when comparing between different approaches

For your case, you might find it helpful to look at the excellent ann-benchmarks.com, and in particular the benchmarks for GIST, which is an image embedding.