iiinova — where your digital self keeps evolving

dcatteeu.bsky.social @dcatteeu.bsky.social · 27/02/2025

You might want to find a leaderboard that ranks the latest models with benchmarks that resemble your use case. Perhaps this one: huggingface.co/spaces/openc...

huggingface.co

Open VLM Leaderboard - a Hugging Face Space by opencompass

VLMEvalKit Evaluation Results Collection

100

dcatteeu.bsky.social @dcatteeu.bsky.social · 26/02/2025

Birds, incl. Goldfinches, don't sleep on their back. This is tonic immobility, a stress reaction where they feign death. petshun.com/article/do-f... youtube.com/shorts/YV8fa... Someone caught these birds, turned them on their back, and filmed them for views and likes.

petshun.com

The Intriguing Behavior Of Finches: Do They Play Dead? | PetShun

Discover the fascinating behavior of finches as this article delves into the topic of playing dead. Learn about the reasons behind this intriguing behavior and gain insight into the world of these cap...

0526

Reposted by @dcatteeu.bsky.social

Professor Gerdes Explains @professorgerdes.bsky.social · 04/02/2025

Hey, I have my YouTube podcast on Spotify too. I would appreciate it if you could like or Retweet (or ReBlueSky) it so that more people can hear about the war in Ukraine. Thank you in advance. open.spotify.com/show/1hgWRnh...

open.spotify.com

THIS is Why Trump Will Support Ukraine

Spotify video

87322

dcatteeu.bsky.social @dcatteeu.bsky.social · 01/02/2025

IMO this is not a sleeping bird. It's experiencing tonic immobility due to stress. hatim.birds's instagram account has several similar videos. He catches the birds, turns them on their back so that they fake death, and makes a video. en.wikipedia.org/wiki/Apparen... www.instagram.com/hatim.birds/

000

dcatteeu.bsky.social @dcatteeu.bsky.social · 30/01/2025

Dario Amodei (Anthropic, an OpenAI competitor) has the same opinion darioamodei.com/on-deepseek-...

Screenshot of a paragraph of Dario Amodei's blogpost: https://darioamodei.com/on-deepseek-and-export-controls

highlighted: "costing substantially less to train" and "DeepSeek's team did this via some genuine and impressive innovations,"

000

dcatteeu.bsky.social @dcatteeu.bsky.social · 30/01/2025

DeepSeek, most probably, illegally used OpenAI models to create training data, but they did not steal or simply copy them to create their own.

000

dcatteeu.bsky.social @dcatteeu.bsky.social · 30/01/2025

2/ Training data is vital to create AI models and just like anyone else in the business they used the best AI models out there to create more and better training data than what's publicly available. OpenAI explains how to do it: cookbook.openai.com/examples/sdg1

cookbook.openai.com

Synthetic data generation (Part 1) | OpenAI Cookbook

Open-source examples and guides for building with the OpenAI API. Browse a collection of snippets, advanced techniques and walkthroughs. Share your own examples and guides.

100

dcatteeu.bsky.social @dcatteeu.bsky.social · 30/01/2025

1/ They genuinely innovated to create and run AI models way more efficiently than anyone else has ever done. That's new and its their invention. It's also what is most important about DeepSeek R1. OpenAI is definitely diving into this as we speak and "steal" this invention. tinyurl.com/mryshhkn

tinyurl.com

DeepSeek's AI breakthrough bypasses industry-standard CUDA for some functions, uses Nvidia's assembly-like PTX programming instead

Dramatic optimizations do not come easy.

200

dcatteeu.bsky.social @dcatteeu.bsky.social · 30/01/2025

IMHO, you were a bit too harsh on DeepSeek. Saying DeepSeek's newest AI model is stolen from OpenAI is misinformation.

100

dcatteeu.bsky.social @dcatteeu.bsky.social · 20/01/2025

Even though Russia has still many hundreds of aircraft left, destroying them has become exceptional. Most notable periods: Feb-Mar 2022: start of the invasion Sep-Nov 2022: liberation of Kherson, Kharkiv 17 Feb-2 Mar 2024: 15 planes destroyed in as many days #Ukraine @professorgerdes.bsky.social

Russian planes (red line) and helicopters (blue line) lost per 30-day period. The numbers drops quickly from more than 100 at the start (February 2022) to about 0 at the end (January 2025) with some occasional upticks.

Year Helicopters Planes
2022 269 283
2023 55 46
2024 5 40
2025 2 0
Total 331 369

Data from: https://index.minfin.com.ua/en/russian-invading/casualties/

010

dcatteeu.bsky.social @dcatteeu.bsky.social · 20/01/2025

Same here, but another way + some UTF □ = big square = 9 △ = big triangle = 1.5 ▵ = small triangle Hypothenuse(△)=√(10) (Pythagoras) ▵ is similar to △ => 1/√(10) = a/1 = b/3 (same proportions) => a = 1/√(10), b = 3/√(10) ▵= ab/2 = 0.15 Answer: □ - 4△ + 4▵ = 9 - 4x1.5 + 4x0.15 = 3.6

020

dcatteeu.bsky.social @dcatteeu.bsky.social · 16/01/2025

The number of drones (Shahed, Geran, and similar) terrorizing Ukraine every day has steadily grown over the past three years. Data from: index.minfin.com.ua/en/russian-i... #Ukraine @professorgerdes.bsky.social

Every year, Russia launched 3× more drones per day than the year before.

Blue dots show the actual reported numbers per day and are most scattered.
Red dots show the 7-day moving average per day. Smoothing out the day-by-day fluctuations they show a clear trend.
A black curve shows an exponentially increasing curve that starts at 3 and grows 3-fold every year. It fits the data over the entire 3-year period.

March 2022: 3/day
March 2023: 9/day
March 2024: 27/day
January 2025: 72/day

1221

dcatteeu.bsky.social @dcatteeu.bsky.social · 12/01/2025

3 As I see it, the bolts are rotated around each other, the lower one first moving towards the viewer. You can as well keep one fixed and rotate only one around the other. This is the same as rotating the other one around its axis in the opposite direction, which makes it move backwards. 🤔

010

dcatteeu.bsky.social @dcatteeu.bsky.social · 08/01/2025

Nice visualization! That's a massive amount of information for very little ink and quite revealing.

200

dcatteeu.bsky.social @dcatteeu.bsky.social · 08/01/2025

Indeed, quite surprisingly, they both grow at 800/year 🤔 1. 4000 < 2000 + 2200 = 4200 2. 4800 < 2400 + 2600 = 5000 3. 5600 < 2800 + 3000 = 5800 4. 6400 < 3200 + 3400 = 6600 ... Let's just hope it's not USD 🤞

010

dcatteeu.bsky.social @dcatteeu.bsky.social · 07/01/2025

Ah, I see your point! Well, apparently not or it would be enforced, no? And I'm wondering, is that for the blind or for AI?

100

dcatteeu.bsky.social @dcatteeu.bsky.social · 06/01/2025

C(n, r) = n! / (r! (n-r)!) = # combinations to choose r items out of n

000

dcatteeu.bsky.social @dcatteeu.bsky.social · 06/01/2025

Another way: 15 ways to draw 2 cards out of 6: C(6,2) = 5x6/2 6 ways to draw 2 non-kings out of 4: C(4,2) = 3x4/2 All draws are equally likely so P(no kings) = 6/15 = 2/5 P(at least 1) = 1 - P(no kings) = 3/5 because there are no other possibilities.

110

dcatteeu.bsky.social @dcatteeu.bsky.social · 06/01/2025

The image shows 6 cards and, IMO the text does not mention the total number of cards. It only states that there are 2, exactly 2, Kings (among the 6 cards you see on the table).

100

dcatteeu.bsky.social @dcatteeu.bsky.social · 05/01/2025

Not my expertise. Would this be the reasoning? The lower pulley provides a 2:1 mechanical advantage. You provide half of the force. The roof kindly provides the other half. The upper pulley is fixed and simply allows you to pull in a different direction. Assuming g=10m/s², 100x10/2 = 500N.

030

dcatteeu.bsky.social @dcatteeu.bsky.social · 05/12/2024

Clear case of PEBCAK For these kinds of questions, you need to let the AI search the web. ChatGPT can do that and, if you ask it, it will get you the 3 correct answers: Clinton, Trump, and Biden. Citations included. As with all tools, you need to know when and how to use them.

ChatGPT answers "Which US presidents pardoned a relative? Please fact check your answers and provide a trustworthy source."

650

dcatteeu.bsky.social @dcatteeu.bsky.social · 18/11/2024

Exactly this: if the will is there.

020

dcatteeu.bsky.social @dcatteeu.bsky.social · 18/11/2024

By actively doing something against it I assume. Are YouTube, Facebook, etc. as infested as X nowadays? I remember Musk complaining about the many bots and wanted some of his money back only to fire the entire moderator department shortly after. It is now clear what those people were doing I guess?

040