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.
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.
It's not ad hoc, although I realize it looks that way, because the guidelines only cover a few of the principles we go by. The rest are described in the endless (and admittedly rather tedious) stream of moderation comments that I put out, which is why I link to HN Search queries so much.
It won't work to add all of them to the guidelines list because that would make the list so long that few would read it, and (worse) it would make it a bureaucratic sort of document that would be most out of sync with the intended spirit. In the long run I (think I) intend to compile those explanations into short (perhaps one paragraph each) glosses on the guidelines, and make it easy to link to.
As for most commenters not discussing the details of the OP: you're right. But that may be too much to expect in this case. To be able to discuss this topic at all without combusting is already a lot. That's also, btw, why I left my pinned comment open to replies (something I never do), and why the meta (why is this on HN, etc.) discussion is not downweighted the way it normally would be.
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.
LLCs do not require rights be assigned fairly to all shareholders if the operating agreement and by-laws say otherwise. This is the case with OpenAI, where the operating agreement effectively makes the fiduciary duty of the for-profit the accomplishment of the non-profit's charter. The pinkish purpleish block of text on the page you linked goes into more detail here.
(Remember, fiduciary does not necessarily have anything to do with money)
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.
Yeah that's my guess too. The claim that he doesn't hold equity always struck me as suspect. It's a little like SBF driving around in the Toyota Corolla while buying tens of millions of dollars of real estate for himself and his family.
It's better to claim your stake in a forthright way, than to have some kind of lucrative side deal, off the books.
For a non-profit, there was too much secrecy about the company structure (the shift to being closed rather than Open), the source of training data, and the financial arrangements with Microsoft. And a few years ago a whole bunch of employees left to start a different company/non-profit, etc.
It feels like a ton of stuff was simmering below the surface.
(I should add that I have no idea why someone who was wealthy before OpenAI would want to do such a thing, but it's the only reason I can imagine for this abrupt firing. There are staggering amounts of money at play, so there's room for portions of it to be un-noticed.)
In recent profile, it was stated that he jokes in private about becoming the first trillionaire, which doesnt seem to reconcile with the public persona he sought to craft. Reminds me of Zuckerberg proclaiming he would bring the world together while calling users fucking dumbshits in private chats.
Oh wow, he's also an effective altruist?! Didn't know that. It's so bad. My take is that there's nothing more hypocritical, and therefore, arguably, more evil than this.
Yeah, although I guess you can read that as: "I will do everything I can to raise the stock price, which executives and employees both hold", then it actually makes sense.
But that $1 salary thing got quoted into a meme, and people didn't understand the true implication.
The idea is that employee and CEO incentives should be aligned -- they are part of a team. If Jobs actually had NO equity like Altman claims, then that wouldn't be the case! Which is why it's important for everyone to be clear about their stake.
It's definitely possible for CEOs to steal from employees. There are actually corporate raiders, and Jobs wasn't one of them.
(Of course he's no saint, and did a bunch of other sketchy things, like collusion to hold down employee salaries, and financial fraud:
The SEC's complaint focuses on the backdating of two large option grants, one of 4.8 million shares for Apple's executive team and the other of 7.5 million shares for Steve Jobs.)
I have no idea what happened in Altman's case. Now I think there may not be any smoking gun, but just an accumulation of all these "curious" and opaque decisions and outcomes. Basically a continuation of all the stuff that led a whole bunch of people to leave a few years ago.
> It's definitely possible for CEOs to steal from employees..
I'm pretty sure that CEO salaries across the board means that CEO's are definitely — in their own way — "stealing" from the employees. Certainly one of those groups is over-compensated, and the other, in general, is not.
What I meant is that there are corporate raids of declining/old companies like Sears and K-Mart. Nobody wants to run these companies on their way down, so opportunistic people come along, promise the board the world, cause a lot of chaos, find loopholes to enrich themselves -- then leave the company in a worse state than when they joined.
Apple was a declining company when Jobs came back the second time. He also managed to get the ENTIRE board fired, IIRC. He created a new board of his own choosing.
So in theory he could have raided the company for its assets, but that's obviously not what happened.
By taking $1 salary, he's saying that he intends to build the company's public value in the long term, not just take its remaining cash in the short term. That's not what happens at many declining companies. The new leaders don't always intend to turn the company around.
So in those cases I'd say the CEO is stealing from shareholders, and employees are often shareholders.
On the other hand, I don't really understand Altman's compensation. I'm not sure I would WANT to work under a CEO that has literally ZERO stake in the company. There has to be more to the story.
> I don't really understand Altman's compensation. I'm not sure I would WANT to work under a CEO that has literally ZERO stake in the company.
This is a non-profit not a company. The board values the mission over the stock price of their for-profit subsidiary.
Having a CEO who does not own equity helps make sure that the non-profit mission remains the CEOs top priority. In this case though, perhaps that was not enough.
Could be that they had an expectation that he not own stock in MSFT since they have such a direct relationship there and found out that he has been holding shares in MSFT.
You have to understand that OpenAI was never going to be anything more than the profit limited generator of the change. It’s the lamb. Owning a stake in OpenAI isn’t important. Creating the change is.
Owning stakes in the companies that will ultimately capture and harvest the profits of the disruption caused by OpenAI (and they’re ilk) is.
OpenAI can’t become a profit center while it disrupts all intellectual work and digitizes humanities future: those optics are not something you want to be attached to. There is no flame retardant suite strong enough.
Worldcoin is scanning people’s irises by having them look into a sphere called a fucking Orb so it can automatically create crypto wallets and distribute global minimum basic incomes after the AI apocalypse.
Altman conceived and raised $115 million for the company.
I could easily see him, or any other insider, setting themselves up administrating a recipient entity for contributions out of those “capped profits” the parent non-profit is supposed to distribute. (If, of course, the company ever becomes profitable at the scale where the cap kicks in.)
Seems like it would be a great way to eventually maintain control over your own little empire while also obfuscating its structure and dodging some of the scrutiny that SV executives have attracted during the past decade. Originally meant as a magnanimous PR gesture, but will probably end up being taught as a particularly messy example of corporate governance in business schools.
Yeah, I agree that the whole legal structure is basically duplicitous, and any attempt to cite it as some evidence of virtue is more emblematic of the opposite...
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.
Just read the abstract and intro for the link you posted - thanks for sharing.
As great as HNSW and other graph-based indexes are, I think PQ and other encoder/decoder-based methods are still incredibly important for ANN search in general. In particular, it should be possible to learn some sort of joint encoding with neural networks targeted towards different modalities.
PQ is just a vector compression/quantization method, not an approximate k-NN search algorithm (like cell-probe or bucketing based methods such as IVF or LSH, or graph-based methods such as HNSW). HNSW can be used together with PQ encoding, for instance.
However PQ does have the nice property that distance computation can be performed with lookup-adds instead of multiply-adds but in practice performance is lower than normal brute-force multiply-add inner product because lookup table gather is slow on CPU or GPU relative to arithmetic. The gain of PQ is by dimensionality reduction (a 128 dimensional vector can be compressed as 8 PQ codes resulting in 8 lookup-adds, versus 128 FMAs for the brute-force inner product, say).
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.
Almost every breakthrough in distributed systems is more recent. LZW is 1980s. Most data storage systems used today had their guts invented in the last 40 years, with only the top layer being older than that.
If you look at "Computer science" as the narrowly-defined field of data structure and algorithm design in a vacuum, maybe things slowed down after 1980, but that's because problems with different constraints just became more interesting.
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.
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...