I once gave a short talk at Harvard. Nearly didn't because I was so intimidated by the crowd. Got through it, hardly able to read the room because of my internal dialogue of focusing on the content, and also convinced I was a fraud. Then I realized I went over my time limit by almost 50%. When it was over, my emotions were such a mess I logically thought I should try to cry or scream to let off the overwhelming negative energy I had built up. It was one unreal bucket of feelings after another. As I collected my clip board and bag, I walked to the side of the hall where I was sure I had seen an exit, someone dodged in front of me and put their hand out and introduced themselves and started asking questions about the talk. Took her card and suddenly there was another. It wasn't until maybe the fourth person when I realized there was a line of at least twenty people. Yeah, stop and say hi. Even with critical questions, complaints. Making a connection with other people is one of the most important things we can do.
This interview with an AWS leader isn’t aging well, from CBS Sunday morning:
Pogue asked, "I don't mean to give anyone ideas, but let's say I figured out that one of these unmarked buildings was an AWS data center, and I blew it up. Are you saying that it's so backed up and redundant that you probably wouldn't notice?" Wood replied, "Yeah, you wouldn't notice. I mean, we might be a bit upset, but you wouldn't notice!"
> concealing that he is a former Apple engineer with direct contacts to the people involved in Apple Silicon development
This is false. I am a former Apple engineer. I did not conceal it (it's the top item of my resume and my LinkedIn and I made my PR from my public Github with my name attached, as opposed to a pseudonym which is expressly allowed by Asahi policy). I also had no exposure, at all, to any internal information or code about macOS, SPTM, or Apple Silicon during my time there.
I also don't have connections to people involved in Apple Silicon development (and I'd add that this does not mean someone cannot contribute, the question is were they exposed to tainted information, which is absolutely not true in my case). I have many friends who work for Apple (I'm a Stanford alum) but none in Apple Silicon directly.
I disagree with the phrasing of this entire thing, but this statement is demonstrably false.
Carney was asking for freedom of movement of goods, people and services, but without full monetary and legislative integration.
So, easier to sell stuff, easier to buy stuff, free to just travel to/from Canada/EU and work there with no extra paperwork, free to buy and sell service from Canada to EU and vice versa.
No Euro, no vote in EU, no ECJ, no payments into each others budget.
Tighter integration than UK, not as tight as Norway/Switzerland/Iceland.
Common defense was already agreed anyhow.
I can tell you this enjoys broad support in the EU public. Although, no joke, a lot of peoples first reaction is: "They need to compete in Eurovision first. Otherwise they should wait for Australia to join first."
This is an excellent use case for completely local, small model inference, yet for inexplicable reasons Mozilla wants to normalize uploading your entire private browsing history to a cloud.
These (Mistral's and Mozilla's) marketing pages aren't candid enough to clearly explain the difference between local and cloud inference, and that they're asking you to consent to enabling the latter. I'd call that the bare minimum of ethics. If you're in a position of authority over your less technical users, you are ethically obligated to give them the full picture of what are you doing and why they should consent.
(Aside to any Mozilla people who might be reading HN, your page here[0] has an oversight, it advertises this model as "Mistral Small 4" but the hyperlink is to OpenAI's model card for gpt-oss-120b).
“81% faster query plans than Postgres”…on an 8 GB dataset that fits entirely in memory, with shared_buffers constrained to a fraction of that, queries warmed before measuring, and read-only SELECTs.
I would be cautious about over fitting, it’s tough to say if those query plans would really be more optimal than Postgres heuristics at scale and with a bit more realistic OLTP workloads.
In any case, such is life with profile guided optimization. Many of us appreciate how database workloads can drift over time and with scale.
Kudos to the author for getting their hands dirty and writing up their experiments.
My first language is Afrikaans, which is a somewhat niche language and hard to find teachers/conversation buddies outside South Africa. (I live in USA now)
I've been using Gemini to live chat in Afrikaans and do impromptu Afrikaans grammar lessons during my solo drives around town. It is phenomenal at speaking the language - like, it really shocks my family members when they hear it.
This is probably the most joy I get from any of my usages of LLMs/AIs. It's been really, really nice getting to speak my language regularly again. =)
So, I'm excited about this release and live chat getting better. I also hope the other frontier labs pick up niche languages like this as well so that I have more options.
Hobby groups projects like this are less fun for a lot of people who used to enjoy interacting with smart people. It's definitely become a game of just spam claude for answers with zero understanding or care for how anything actually works. That's fine to get things done and fine for a lot of side projects, but it definitely ruins the joy that people have in understanding systems and working with intellectuals.
Researchers asked frontier models to play chess. Have a look at the MAR rates in Table 3. When not explicitly told which moves were legal, no model identified legal moves at a rate better than 80%. Many asked for more illegal moves than legal moves. And even when explicitly told which moves were legal, the models continued to ask for illegal moves. With illegal asks discarded, none of the bots could beat a chess model calibrated to 1100 ELO.
The author of the originating post says that "current frontier models need laborious oversight and guardrails on even the simplest tasks", and he's absolutely correct.
> I strongly dislike CUDA. Once you have allowed that proprietary cr*p
Genuine question...why not just type "crap"? It's not even that much of a curse, but I've never really understood the point of self-censorship. If you don't want to curse then you could just use a non-curse word.
MAYBE operating systems shouldn't have gatekeepers which can deny access to billions of customers for any and no reason at all. Apple and Google are WELL past due for regulation in this space. The fact that we cannot download and install software from the internet onto our phones JUST like we can do with our computers is a symbol of our inept and ineffective politicians.
And before someone says "well akshully you can technically do it on Android," Google has been working tirelessly to make it as onerous as possible for both developers and customers. For example, users need to separately authorise each app (browser, files manager, alternative store, etc) to install an APK from outside the Play Store. Google also does background scans using "Play Protect" which will periodically delete apps Google doesn't approve of. This happened to me with SmartTubeNext. I have a dozen other ways Google ensures users are discouraged from stepping outside the Play Store.
I know HN readers and posters just read numbers and can't be bothered to read, but please read the methodology before making any claims.
> About their ELO ratings from their own website:
> A field-relative rating calculated within ChessBench. It compares performance among the tested models and is not a direct equivalent of a human chess rating.
I am around 1600 elo in over the board I can mop up Astra Fable etc even if I give them literal infinite time and all the subagents and internet access..
Please folks at least use your AIs to read stuff before making claims.
AI is not GM level, it's not even 1600, I am 1600 by using memorized openings people frequently fall for with very basic intuitions.
A GM is 2600 they can beat me in under 20 moves...
Why do I even scroll through this website. For a moment I truly felt fooled, but then I read like a human should.
Maybe I should stop doing that will be a happier life, don't think just believe in the AGI.
No idea what this actually means, if it means easier trade, I'm all for anything that reduces our dependency on the yanks, as long as we don't need to be involved in their regulatory or legal regimes. The Americans are no longer our friends, Salut, mes amis Européens. :)
This is really clever from Apple. The journalist use case is just the PR story. This will be really useful for identity verification and insurance apps, and has the potential to shift from "you need a smartphone to be able to live normally" to "you need an iPhone to be able to live normally".
There are already plenty of insurances that require you to submit claims through a smartphone app that tries to essentially do this by capturing sensor metadata etc. - those don't need to be nation-state resilient, just Joe the Crackhead Insurance Scammer resilient, so this works. Likewise, more and more things online require identity verification (either officially or disguised as age verification).
Edit: And while "a nation state actor can spoof this" is a problem for the journalism use case, the insurance/ID verification use cases are perfectly fine with anything that raises the bar but could be bypassed with enough effort. Also, the journalism use case suffers from the same fundamental issue all of these use cases suffer from: People will "verify" the picture by looking at the repost of a screenshot of the verification UI, not by verifying the original themselves.
> Gemini 2.5 Pro, O3, Claude Sonnet 3.7 and ChatGPT 4.1
The gap in capabilities between those models which they tested, and actual current frontier ones is enormous. I would not trust that any conclusions they made are applicable.
Dear fellow humans from "Hacker News". Hacking a driver that in itself documentation to black box Apple hardware is not any different from hacking $10 4G LTE modem.
Fact that a person who was not previously driver developer can achieve this in a few weeks is pure wonder. No matter what tools are used.
Leave legal questions to Linux Foundation laywers or whoever is responsible to accept or not accept the patches.
If Apple actually wanted to prevent any of this from happening they can just lock bootloader. They not just gonna do it now because someone used fancy text generator to make working GPU driver.
Attacking a person who spent their personal time and money on making walled garden black box hardware more usable, more privacy friendly and further away from landfill is a crime against the future of humanity.
It doesn't read as emphasis to me. It reads like the person is trying hard not to curse, and they think "crap" is a curse word. It's a little bit adorable, like I'm reading a comment from an obedient child.
I strongly dislike CUDA. Once you have allowed that proprietary cr*p into your C++ codebase, it is very hard to get rid, and you end up with code that is either tied to a single vendor or an #ifdef hell, probably both.
The best way to program GPUs is face up to the reality that they are not the same machine as the CPU, write your kernels in separate files, and launch them manually, like in Metal, OpenCL, and D3D12, etc.
These days we even have DSLs like Triton that make kernel writing much more ergonomic than anything you would hope to achieve in Rust.
I'm partly to blame. I accidentally ran an illegal unregulated money transference service over Google Play.
It let users cash out their Google Play Credits for real cash, which I automatically wired them.
Someone then hacked in to a major bookstore chain, stole piles of Google Play Gift cards, activated them using their access, and used my app to get cash for them.
Luckily, the whole thing blew up on me before I got in serious trouble, rightfully so, and the app was removed by Google, then an investigation followed. A ton of copycat apps popped up immediately after, then a few months months later Google announced their app review process.
Related to the point about worse performance, I'm pretty sure I was there when reduce was "banished" from Python 3 -- demoted to functools.reduce(), instead of the builtin reduce() in Python 2
The story is that sometime in 2006 or 2007, Guido van Rossum was debugging why a web page in Google's internal code review tool (which he wrote) was taking 30+ seconds to render.
This is basically a "production" incident, since thousands of Google engineers relied on the tool. Requests like this were probably tying up threads and exhausting thread pools, perhaps
Eventually it was tracked down to a line wrapping algorithm written with reduce(). I don't think he wrote it -- it may have come in through a dependency. As many know, reduce() is basically:
s1 + s2
s1 + s2 + s3
s1 + s2 + s3 + s4
...
And that's O(n^2) when s_i are strings. And I think it showed up if you viewed a 5000+ line diff, or a 5000+ line file. (Newer programs like Github also suffer here)
I believe, in Python at that time, += was already optimized to avoid this (just like essentially all JS VMs are). Or you can use the idiom of append() to list and join() after.
But reduce() basically forces the inefficient implementation, and I'm sure this is still true in Python 3.
---
So basically Guido spent a long time debugging a performance problem related to reduce(), and made the decision to eject it, to help users avoid "footguns". I was his officemate at the time, so I recall this, but I wasn't involved directly
Also, somebody contributed reduce() to Python way back in the 90's, as well as other functional idioms. He wouldn't have added that himself -- it was never his preferred style.
He preferred a more imperative style. But he allowed those contributions, and then slightly regretted it later.
With Signal, the biggest issue we have is that review times are extremely inconsistent. Sometimes it's 4 hours, sometimes it's 5 days, and there's no visibility as to why. Our working theory is that there's automated and manual queues, and occasionally, for whatever reason, we fall into the manual queue. But when you work on an app that has weekly updates, randomly getting hit with a review time of several days really throws off your groove. And it can obviously be terrible for moments where you're fixing a critical issue.
