Tankenstein

Most of the time when upgrading models we have needed to change prompts to get the same performance (let alone better performance). Usually, your prompt is overfit to the specific model doing the specific task. For example often your previous prompt is overspecifying and creating contradictions that a dumber model would just gloss over whereas a smarter model will try even harder to follow.

Render.com has a similar value proposition. I’ve used them and am pretty happy. Railway seems to have more bundled observability built in, that i’d like in render.

I moved to render years ago and have been very happy with the decision. It feels like heroku, if it never got acquired by salesforce and kept improving.

Happy new year from Estonia!

Looks like a demonstration that using `long double` math requires dipping into x87 instructions, specifically the `fldt` instruction: "floating point load ten bytes".

This is the first time since GPT 4.1 that I think I can upgrade our main agent model. Any noticeable amount of reasoning has been too slow for us, since the model is having a real-time conversation with the user. "minimal" reasoning GPT-5 performs terribly, it's significantly dumber than GPT 4.1 in a long, multi-turn conversation with tools.

This time, I just dropped it in and at first glance it seems to work well. I'll probably upgrade over the weekend if I see a boost in performance somewhere after tuning the prompts.

x86 to arm compatibility layer they are using to run windows games on the machine/frame

A few additional points of context to consider:

- The "Turkey taught Putin" narrative is kind of BS. The leader of Turkey imprisoned the pilot who shot down the Russian jet, essentially saying he acted on his own. Afterwards, Turkey bought Russian military equipment. This reaction is the opposite of sovereignty.

- Speculation: Russia is looking to provoke Europe into responding. The thinking is to get Europe to focus on their own air defense, over assisting Ukraine. This is why the general, and correct, response in Europe is to keep calm and keep sending air defense equipment to Ukraine instead of freaking out. There of course are some lines where it is not possible to shrug it off, like recently in Poland, where the situation was so unsafe that we had to react with weapons. The most important element of Europe's security is that the Ukraine war does not end in a victory for Russia.

Edit: Formatting

Thank you for the additional context. Based on this, I agree that Oslo might not actually be related.

While the motive is speculation - each event the poster pointed out happened. These are not normal events.

If the intent is to harass - lights on is the way to go, to make their presence very clearly known.

There's footage of this incident, footage of monday's Copenhagen incident, and there were arrests in Oslo on monday for another incident over Oslo.

When this happened on monday night at Copenhagen airport, they described them as "big". Speculation: seems like large commercial quadrotors.

Something quite cool they shipped with the 26 family of operating systems is a new system-level API to let developers access the built-in on-device models directly: https://developer.apple.com/documentation/FoundationModels

This person posts large amounts of misinformation, best not to read into it.

Foldables need to have very thin construction as when folded, you essentially have two phones on top of each other. There's speculation online that the air was launched off the back of their internal foldable R&D.

The pixel has had a "visor" style bump for many years, last year's model rounded it off for the current look. This year's model is design-wise very similar to last year's.

It relies on cameras.

Wikipedia: The flush cockpit means that the long and pointed nose-cone will obstruct all forward vision. The X-59 will use an enhanced flight vision system (EVS), consisting of a forward 4K camera with a 33° by 19° angle of view, which will compensate for the lack of forward visibility.

Bring more must-have parts of the stack into the offering:

- Object storage

- Metrics (managed prometheus + grafana? or preferrably your own lightweight version, like your log system, which mostly gets the job done)

Thank you, I wasn't aware of China using metres. It turns out Russia uses them as well, confusingly below the transition level.

Sorry, not a native speaker, I was under the impression that aerospace means air and space. I guess i meant aviation.

I didn’t imply knots are mph, I used the slash to signify “or”. They are completely different units, but both are used. Sometimes the airspeed indicator even has two scales, one for kt and one for mph.

Just a guess, but aerospace generally works with feet for altitude and knots/mph for airspeed, internationally. I’m doing a PPL in Europe and we, like everybody, use feet and knots/mph. I believe this is because the US have been on the forefront of aerospace regulation (a set of rules called the chicago convention is the basis of all air law) and aircraft manufacturing.

Presumably, Elon would leave otherwise. Looks like the majority of tesla shareholders would like Elon to stay.

Makes sense, your argument is clear now, my mistake.

The original commenter mentioned 70B, not 7B.

Many startups have started over the past few years, trying to build infrastructure (shovels) for companies to integrate LLMs, or specific chat copilots trying to cater to a specific usecase. Most are dead in the water once OpenAI subsumes their feature set.

Rooting for these folks. Very happy that they're building mistral at this pace from Europe – breaking the narrative.

Wise lets you access your account and money with API-s. You can create a personal token (and webhooks if you want) at https://wise.com/settings and you're off to the races. Api docs are at https://api-docs.wise.com

Disclaimer: I work on Wise. (We're hiring: https://wise.jobs)

I find their approach of end-to-end machine learning absolutely fascinating. Since the video is very long, I'll explain the gist: instead of the traditional approach of labelling everything in the scene using neural nets and then trying to divine a driving plan based on this information, comma is trying to behave more like a human and going straight from perception input to neural net to car controls output (expressed as a path).

This to me is a fascinating approach and raises interesting questions about the observability of the system and how much we can trust systems like this, where validation is based on testing of the behaviour of the whole system rather than validation of its subsystems.

I believe that comment was directed towards posts about price fluctuation of existing public companies, while this post is about the company going public. Previous listing posts have led to substantive discussion in my experience.

Disclaimer: I work on wise.