ricardobeat

7 ms·

1.
by ricardobeat 5d ago
This page has been in 'New model — collecting baseline data. Degradation detection paused.' state for months now. It seems to never say 'degraded'. If you look at the graphs, the latest benchmarks are showing a pretty si
2.
by ricardobeat 5d ago
They increased yearly subscription just two months ago, from €139 to €159. Those €20 alone should pay for the ads. And Disney+ is already by far the most expensive subscription of all, while providing very limited content. This is completel
3.
by ricardobeat 6d ago
“byte-identical” is one of Claude’s favorite expressions
4.
by ricardobeat 6d ago
A little hard to interpret as the Huawei pictures are not cropped to the same frame. When zooming in the difference is less striking.
5.
by ricardobeat 6d ago
It’s easy to say, but do you want a teenage mind to compete against an army of thousands of marketers and engineers, with billions of dollars on the line, to see who wins that battle? It’s not a level playing field.
6.
by ricardobeat 6d ago
> Teens will find creative ways to access these or even more unregulated spaces The article is missing the point entirely. The conversation is not about “regulated spaces”, but removing them from spaces created entirely to drive advertis
7.
by ricardobeat 6d ago
> and to date there seems to be no appetite for legal enforcement of these obligations. Three paragraphs earlier, the virtues of openly sharing code were being sung. > while if I keep my code closed it is far more difficult to find th
8.
by ricardobeat 6d ago
Don't waste your time - the 'meltdown' was the underlying LLM starting to speak Chinese while being interviewed. Their character has slow reactions, seems quite dumb and has no personality at all. They are probably running an
9.
by ricardobeat 7d ago
What linux distro/desktop do you consider more user friendly than MacOS? The best experience I’ve had is with Omarchy 4, but it is not appropriate for average computer users.
10.
by ricardobeat 7d ago
That’s statistically unlikely (to not say impossible), isn’t it? Plus the compute required to brute force a key is not available at inference time.
11.
by ricardobeat 8d ago
Tells you something about what Apple/Samsung/etc are able to achieve with their manufacturing and supply chains.
12.
by ricardobeat 8d ago
These optimizations are impressive, but it gets me thinking: at what point does a company become a collection of impenetrable siloes, where nothing really does what you expect? Maybe know with AI this is less of an issue as exploring a code
13.
by ricardobeat 8d ago
I like efficiency! Nowadays my phone and laptop are constantly running out of storage since every calculator app is 1GB.
14.
by ricardobeat 8d ago
Is there a downloadable sample somewhere? Would like to see how large a native build is.
15.
by ricardobeat 9d ago
Three stacked skills + a 400-line python script is not within my definition of “very easy”, but I appreciate the effort!
16.
by ricardobeat 9d ago
What does that mean for that donation, and why is the wallet still listed on the website?
17.
by ricardobeat 9d ago
The documentation is entirely about how to use it, how to manage the skills and configure it. What I'd like to see is what it is, how and why it works, ideally backed by some benchmarks. Otherwise it looks like just another pile
18.
by ricardobeat 9d ago
In my experience Opus and Sonnet 5 subtly ignore most instructions related to writing style, and continue to sound the same half of the time. Do you have a successful skill/prompt to share?
19.
by ricardobeat 9d ago
If you're thinking of the UI style, definitely not Claude. It is incapable of writing a clear sentence like "what each step's samples are made of", would have used all-caps for everything, more padding and gradients.
20.
by ricardobeat 9d ago
They both are in the 50-100 tok/s range. The Mimo v2.5 Pro Ultraspeed beta could reach 1000 tok/s, hoping they can do something similar for the new model, it was amazing.
21.
by ricardobeat 9d ago
For reference, Mimo-v2.5-Pro scored 19% on DeepSWE 1.1. This is looking great. Fable scores 70%, Kimi K3 69%, Astra 74% (all on max effort). https://deepswe.datacurve.ai/blog/deepswe-v1-1
22.
by ricardobeat 10d ago
A bit meta , but this is one of the first posts in Anthropic's official blogs I've read that has distinctively Claude-style prose, thought less prominent than usual: What Claude makes doesn’t need its own place either.
23.
by ricardobeat 11d ago
The doom demo shows the program state / query.
24.
by ricardobeat 11d ago
I've just started working on 3D modeling software of my own, though this looks to be far ahead. Mostly focused on 3D printing - creating printable models is pretty expensive and tedious at the moment, and fully native. [1] https:&#x
25.
by ricardobeat 11d ago
You think farmers are willing to risk the murder of hundreds of their fellow countrymen as a form of protest? That’s plain terrorism.
26.
by ricardobeat 12d ago
Not everything should have a dark mode. As a style that requires bright contrasting borders and heavy shadows, this is not particularly pleasant as a dark theme. See the “Hardbrut” in dark mode, posted elsewhere in this thread. The white “s
27.
by ricardobeat 12d ago
Those two websites have nothing in common. “Neobrutalism” might be wrong in concept, but it was coined around 2023, that ship has sailed.
28.
by ricardobeat 12d ago
At current hardware prices, this is worth buying to use as a server!
29.
by ricardobeat 12d ago
And this is why we'll have to live with AI slop for the next decade
30.
by ricardobeat 12d ago
May I suggest removing/rephrashing the "you are not the ship" subtitle. It's obvious AI "this button is not red" phrasing that cheapens it a bit.

More ›