Mike Ushakov (@ya_mike) header
Mike Ushakov (@ya_mike) profile picture

Researcher at Leiden University and founder at https://teapaw.org/t.co/cVa9KtI1du. Science, startups, music, and dogs.

Amsterdam, The Netherlands
Joined March 2009
Photo by Mike Ushakov (@ya_mike)Photo by Mike Ushakov (@ya_mike)Photo by Mike Ushakov (@ya_mike)Photo by Mike Ushakov (@ya_mike)Photo by Mike Ushakov (@ya_mike)Photo by Mike Ushakov (@ya_mike)Photo by Mike Ushakov (@ya_mike)Photo by Mike Ushakov (@ya_mike)Photo by Mike Ushakov (@ya_mike)Photo by Mike Ushakov (@ya_mike)
1/11 Can you pet the dog in an AI-generated video game? Rarely. I made DogLM — an LLM benchmark to measure it. My findings are in this thread. @CanYouPetTheDog
2
1
2
10
1,362
It is fun to learn that "Digger" flopped at the US box office. The best Iñárritu movie I've seen.
54
Jon Stewart's interview of @hilbertspaess is the single best TV piece on the AI race in the whole television history. I really hope the people working in the labs are listening. Go out and hug some humans, the world is not only about a circle jerk. youtube.com/watch?v=lnLDjXy2…
1
124
1/4 New models were released last week, so an update: GPT-6 Astra generates the nicest-looking dogs so far and many of them are pettable, but Gemini 3.7 Flash makes more interactive dogs in total and keeps the #1 🏆. The dogs in the video are from Astra 🐶. Full results are below
1/11 Can you pet the dog in an AI-generated video game? Rarely. I made DogLM — an LLM benchmark to measure it. My findings are in this thread. @CanYouPetTheDog
1
1
1
7
435
3/4 Muse Spark 1.3 is a VERY capable model to generate interactive dogs: it scored slightly above all Anthropic’s models while also being one of the cheapest models per game in the whole experiment (15 times cheaper than Fable 5.1).
1
56
Also, many thanks to @TheZvi for mentioning my DogLM benchmark in 'Don't Worry About the Vase'. Probably not the best timing for half-serious benchmarks, but it is also never a bad time to pet the dog.
1/11 Can you pet the dog in an AI-generated video game? Rarely. I made DogLM — an LLM benchmark to measure it. My findings are in this thread. @CanYouPetTheDog
2
123
10/11 I’m confident that we need more benchmarks that try to capture AI systems’ spontaneous ability to be warm and humane. I explain the idea of the benchmark and the first results in my blog: mikeushakov.com/machines-of-…
1
3
96
11/11 I also want to say a public thank you to my favorite X/Twitter account of all time, @CanYouPetTheDog. You’ve made a lot of people smile.
1
85
Why should we build flying cars and travel to Mars if all that can be done in virtual reality, right? #Metaverse
1
5