yathaid
They are not informal AFAIK - https://github.com/PGPerfFarm. A lot of these heavy duty OS projects have explicit Perf testing/nighlites (see Lucene's at https://benchmarks.mikemccandless.com/)
If you are saying the PG regression suite doesn't cover these perf tests - that is fair. I consider perf to be part of the "regression framework" generally.
>> How do you know your new tests are right? That the new behavior is right?
This is a more generic question in the LLM world. Reliable verifiers is what drives LLM loops. If you don't have these, you have no idea if you are actually making progress or you just generated code that returns 42 for every question. You needs something to actually ground the LLM output against. For existing features, the reliable verification is the existing regression/perf suites. For new features, the regression/perf suites should expand to fit. There are ways to go at this that range from ad-hoc (line/branch coverage) to fully formalized (verification-aware languages like Dafny).
You mean the ones encoded in the regression suite? I think your argument is valid in many medium-to-faang firms where the application is going to have encoded business logic that isn't explicitly tested for. If anything, projects like PG are the exact opposite: no business context and regression suites that test every possible scenario due to the accumulation of bug fixes and context over many years.
>> Run it in production for a year in a sufficiently large system and you might be somewhere.
How do you know every release of PG doesn't break in the setting of "a year in a sufficiently large system"?
Evidence: see the amount of nonsense in the postgres rewritten in Rust story - https://news.ycombinator.com/item?id=48841676 where there are two contradictory claims made:
"postgres is so stable I will never trust a rewrite."
"covering 100% of postgres regression suite doesn't guarantee you have replicated every behavior."
Folks who are okay with LLMs think the regression test suite is the spec and is the guarantee of stability. How else can it be? If you are depending on some behavior not covered in the regression tests, how do you know the next minor release won't break you?
Folks who are against seem to imagine a platonic ideal of PG which conveniently is the original PG implementation by tautological definition. So no rewrite can ever meet their bar.
You can already imagine Anthropic working with a bunch of shady brokers to "remedy" this situation.
Please do read the full PR, it is informative and funny!
>> My workflows involve opening and closing up to hundreds of terminal or tmux tabs a day.
What?!?
Without an explanation of what they author is calling out as flaws, it is hard to take this article seriously.
I know engineers I respect a ton who have gotten a bunch of productivity upgrades using "AI". My own learning curve has been to see Claude say "okay, these integration tests aren't working. Let me write unit tests instead" and go on when it wasn't able to fix a jest issue.
__need__ is doing a lot of work here. There is no forcing function to get OEMs to do this ASAP: 1) the market doesn't really care that much 2) there are no regulations around this (and even if they were, can you immediately recall a tech exec going to jail for breaking the law ... )
good is doing a lot of heavy lifting here. The point of TLA+/Pluscal is to have a proof of the soundness of the design.
This highlights one of my personal bugbears. People have a mental barrier when it comes to recurring, low-cost payments; even though the net sum is small in comparison to other things that they wouldn't think twice to pay for.
A $5 latte every workday comes to (260 * 5) $1300 annually. Obsidian sync is $96. Why would you not pay this amount for a tool you use everyday?
If anyone else was searching for the dataset, it is at https://huggingface.co/datasets/AvaLovelace/StableText2Lego
It contains " contains 47,000+ different LEGO structures, covering 28,000+ unique 3D objects from 21 common object categories of the ShapeNetCore dataset".
Local inference instructions are over at their github page - https://github.com/AvaLovelace1/LegoGPT/?tab=readme-ov-file
Yes. In this case, it is the artist's sole right to reproduce said images, based on their creative output.
>> decreases scarcity, it doesn't increase it
What does scarcity have to do with stealing? You can steal bread and reduce food scarcity, but that is still theft.
[1] - https://www.inc.com/bruce-crumley/doge-cost-cuts-zero-in-on-...
I have theorized some root causes:
- They cannot differentiate between well-meaning friends and high quality information i.e. there is a fallacy of "this person is honest, hence this forward they just sent me is true".
- Starting from at least my generation (born in late 80s), there is an understanding of "echo chamber effects", personalizing newsfeeds for engagement etc. There is some inoculation against content meant to trigger/resonate with specific sub-groups. I have found this to be completely lacking in discussions with my parents/their generation.
All these make it hard to move them out of the dis-information locus they fall into.
>> But in the limit of tokens generated, the chance that they generate the correct answer still decays to zero.
I don't understand this assertion though.
Lecun's thesis was errors just accumulate.
Reasoning models accumulate errors, track back and are able to reduce it back down.
Hence the hypothesis of errors accumulating (at least asymptotically) is false.
What is the difference between "Probability of correct answer decaying to zero" and "Errors keep accumulating" ?
In the previous paragraph, the author makes the case for why Lecun was wrong with the example of reasoning models. Yet, in the next paragraph, this assertion is made which is just a paraphrasing of Yecun's original assertion. Which the author himself says is wrong.
>> Instead of waiting for FAA (fully-autonomous agents) we should understand that this is a continuum, and we’re consistently increasing the amount of useful work AIs
Yes! But this work is already well underway. There is no magic threshold for AGI - instead the characterization is based on what percentile of the human population the AI can beat. One way to characterize AGI in this manner is "99.99% percentile at every (digital?) activity".
But, you can find this attitude pervading the whole linux desktop ecosystem. This post may as well be titled "Why it will never be the year of the Linux Desktop".
KANs have no advantage in terms of computability. Why are they a promising pathway?
Also, the splines in KANs are no more "explainable" than the matrix weights. Sure, we can assign importance to a node, but so what? It has no more meaning than anything else.
There are only two objective measurements needed:
-is it making progress towards its goal?
-is it able to acquire capabilities it didn't have previously?
I am not sure if even the first one is objective enough.
Dismissing the argument without stating why you aren't convinced just comes across as a form of AI ludditism.
That is a big accusation there. Do you have anything like evidence or just your hurt feelings?
An oversight committee actually doing oversight?! Burn them at the stake.
>> force them to pay for expensive defense in court
Gee, will nobody think about the poor billion dollar companies and their stranglehold on various sectors.
>> then losing when the court decides they did nothing illegal
In an article whose point is a judge allowing the case to move forward.
Sir/Madame/Person, your shilling is top-notch. To the obviousness of such, I doff my hat.
I tried reading what redstone blockchain was but I cannot make heads or tails out of this:
>> Redstone is a chain built from the ground up for onchain games and autonomous worlds.
>> On Monday, Wood told Brinkema that Google intentionally put itself in this position to "manipulate the rules of ad auctions to its own benefit," The Washington Post reported.
A better way to understand this is that Google allegedly forced publishers to use Google's ad server and ad exchange if they wanted traffic from Google ads. From [1]:
>> After buying DoubleClick, Google tied its control over advertising demand to publisher use of its software. As the DOJ put it in the complaint, "If publishers wanted access to exclusive Google Ads’ advertising demand, they had to use Google’s publisher ad server (DFP) and ad exchange (AdX), rather than equivalent tools offered by Google’s rivals." The result is that it acquired a monopoly across the entire industry, in the software publishers use and the matching engine for advertisers.
[1] - https://www.thebignewsletter.com/p/a-post-google-world