Is there any article explaining how AI tools are evolving since the release of ChatGPT? Everything upto MCP makes sense to me - but since then it feels like there is not clear definition on new AI jergons.
Source: EPFL, ETH Zurich, and the Swiss National Supercomputing Centre (CSCS) has released Apertus, Switzerland’s first large-scale open, multilingual language model — a milestone in generative AI for transparency and diversity. Trained on 15 trillion tokens across more than 1,000 languages – 40% of the data is non-English – Apertus includes many languages that have so far been underrepresented in LLMs, such as Swiss German, Romansh, and many others. Apertus serves as a building block for developers and organizations for future applications such as chatbots, translation systems, or educational tools. The model is named Apertus – Latin for “open” – highlighting its distinctive feature: the entire development process, including its architecture, model weights, and training data and recipes, is openly accessible and fully documented.
Apertus is available for commercial use based on its license. It doesn't produce state-of-the-art (SOTA) results, but for many organizations, it greatly reduces risk of copyright infringement and even if it does, there is a direct way to address it. In fact, if you were posting in good faith, I would expect people very concerned about questionable training data to be more aware of Apertus.
Simple in the sense that it would have been quick to do, yes, and I almost did that instead. But I felt it would have been more duct tape on an increasingly fragile and outdated system, and I decided the best path forward for the project was to modernize the whole system instead.
> At that point, I reached for an age-old tool that has gotten more useful in the modern age: binary search. That is, you explain the symptom to your coding agent. Then you have it repeatedly remove stuff from your code that might be causing the problem
Can someone give me some high level pointers on how to setup this scaffolding?
When I read the first sentence, I expected the author to use `git bisect`.
However, what the author seems to have done is used a prompt with claude that probably looked something like this:
"Some piece of code is causing the page to load very slowly. To debug this, I'd like to use binary search, where we keep commenting/uncommenting 50% of the remaining code, and then I manually check if the page is still very slow. Let's start now; Comment out a component (or parts of a component) that you estimate is 50% of the page, and I will tell you if the page is still slow."
> When I read the first sentence, I expected the author to use `git bisect`.
Exactly the same. I'd first (or have Claude or whatever) use git bisect. Then I'd focus on the commit that is problematic. And why not have the "binary" detective work TFA talks about.
That said... 1600 ms per heart emoji. Unicode strikes again, combined with a fallback to SVG because that font couldn't color a heart emoji. What a rube-goldberg of a kitchen sink world we live in. It's just sad really.
Honestly beam me back up to the late 80s or early 90s: give me a car with pop up headlights, a Hi-Fi stereo with a CD player, an arcade cab, a Commodore Amiga and my ASCII charset. Life was better.
How well agents can do this is mostly proportional to how well they can understand and navigate your codebase broadly.
There are various contributing factors to this, but they include clear docs, notes and refactors that clear up parts the agent commonly gets confused by, choosing boring technology (your dependencies are well understood) and access to command-line tools that let it lint + typecheck + test the code. A lot of the scaffolding and wiring necessary are built into Cursor and Claude Code themselves now. Hope that helps!
I do this all the time in a dumb but effective way. Add logging statements to code paths that drop timing info. Another dumb but effective way, instead of using a step through debugger, is drop "here, value is {val}". Telling claude to do this is trivial, it's quick, and it can read its own output and self-solve the problem all with just the code itself.
IMHO git bisect is slower, especially depending on the reload/hot-reload/compile/whatever process your actual app is using.
Alex Honnold: No free soloist ever died doing anything cutting edge. Nobody died doing something really hard. A handful people died doing things that are easy. Most soloist died in different types of accidents...base jumping, rogue wave.
Factually inaccurate, the Wikipedia page for free solo climbing has an entire section on prominent free solo climber deaths.
In my view this guy is pretty irresponsible especially for promoting a “sport” that is unnecessarily dangerous in the most preventable way imaginable.
A bunch of kids and stupid adults watched that live stream and a non-zero amount of them now think they can try the same thing without anywhere near as much training and skill.
Not gonna lie, this kind of extreme sport doesn't attract the brightest people.
I have a fascination with this kind of stuff. Taking risks is sometimes necessary and worthwhile in life, but seriously, risking your life just so you can say you realized a specific feat is really dumb.
Especially when we have the technology to do the thing just fine with no problem whatsoever.
I think those people live too comfortable lives and try to feel something. They could apply those skills to some meaningful work, but of course it doesn't get the same recognition.
In his El Capitan climb (Free Solo), Alex was worried about cameras or presence of friends watching interfering with the climb. As oppose to that, this climb must have felt very different!
>Climbing star, 23, dies after falling from Yosemite's El Capitan [this past Wednesday]
>Balin Miller, 23, was live-streamed on TikTok ascending and subsequently falling from the monolith on Wednesday.
>Details of what caused the incident are not clear, but Miller's brother Dylan told AFP he was lead rope soloing - a technique that enables climbing alone while still protected by a rope - on a 2,400ft (730m) route named Sea of Dreams.
>He had finished the climb and was hauling up equipment when he likely rappelled off the end of his rope, Dylan said.
>Tom Evans, a Yosemite-based photographer who witnessed Miller fall, told Climbing magazine he called 911 after Miller tried to free his bag, which was stuck on a rock.
Not sure what your point is here. He died rappelling when his bag got stuck. He didn't die free soloing. And it's unrelated to Yosemite. It was an avoidable accident and very sad. Climbers tend to die from Rappelling more than anything else. And, it's completely incomparable to Honnold's recent climb.
Yes, Freerider (the route he climbed on El Capitan) is much harder than the climbing on Taipei 101. The style of climbing is also very important, some of the moves on Freerider are very insecure and hard to climb in a reliable way, whereas on Taipei the difficulty largely comes from doing the same moves over and over again which means your body gets tired in a specific ways.
The climbing on Taipei was way more chill for him than the climbing on Freerider.
I'd assume - unlike El Capitan - the pitches here are all pretty much identical, so by the time he got to the third or fourth floor he had it figured out
If you watch the climb you'll see that the skyscraper definitely wasn't quite so straightforward - there were some interesting challenges along the way.
Of course, no question El Cap was technically far more challenging.
I wondered the same because there was a helicopter the entire time. Also the very tip top of Taipei 101 appeared to have many cameras mounted on it, at least through binoculars.
reply