vunderba

37 ms·

811.
by vunderba 8mo ago
Yeah I think that it's part of the issue with a single "squashed" comparative metric. Some users are going to grade higher based on the overall visual fidelity and others are going to value following the prompt. For a point o
812.
by vunderba 8mo ago
For reference, have you seen the Artificial Analysis Image Arena Leaderboard? They also show you two images from anonymized models (shows after you vote), and calculates crowdsourced ELO ratings. https://huggingface.co/space
813.
by vunderba 8mo ago
https://archive.ph/w5qyD
814.
by vunderba 8mo ago
Love the lemongrab. Also use a number of Adventure Time related pictures (Peppermint Butler) on my site. I always add easter eggs to my profile/about pages - the number of people that discover them is relatively small but it's rew
815.
by vunderba 8mo ago
Great article! It’s been a while (obviously), but I always used to compile the QB-related runtime modules (QB 4.5) directly into the EXE itself, and the size of most of final binaries fit comfortably on a 1.44 MB floppy. We later had a copy
816.
by vunderba 8mo ago
Ah yeah that'll get complicated quickly. :( It'd be nice if stockfish or its variants exposed the "ply searcher" with some kind of a callback so you could quickly kill a ply based on your custom constraints. When I was d
817.
by vunderba 8mo ago
Understood. Research would make me especially leery; I’d be afraid of losing any potential gains as I'd feel compelled to always go and validate its claims (though I suppose you could mitigate it a little bit with search engine tooling
818.
by vunderba 8mo ago
Nice, had no idea they'd ported it over to WASM. I've used the original Fairy to play Chinese chess in the past though and it worked great!
819.
by vunderba 8mo ago
I can't speak to using local models as agentic coding assistants, but I have a headless 128GB RAM machine serving llama.cpp with a number of local models that I use on a daily basis. - Qwen3-VL picks up new images in a NAS, auto capt
820.
by vunderba 8mo ago
Use a local model such as Gemma3 with a prompt such as "strictly limit changes only to spelling issues, syntactical errors, and punctuation." That way, it's basically functioning like Grammarly on steroids. Asking an LLM fo
821.
by vunderba 8mo ago
> The only limitation is that its internal search tree imagines unconstrained future moves. Yes this is what I'm saying. Evaluation can be way more than just "slightly off" though and can make the evaluation MUCH weake
822.
by vunderba 8mo ago
> That function runs client-side and filters the list of legal UCI moves before being passed to Stockfish's searchmoves parameter, so Stockfish's skill level, depth, and randomness all still apply correctly within the allowed
823.
by vunderba 8mo ago
Whenever a friend complains about how much they hate games with boss rushes (a mechanic to artificially inflate the length of a game by forcing the player to re-fight all the previous bosses from previous levels in order) - I refer them to
824.
by vunderba 8mo ago
Nice job. I have no point of comparison (having never actually used it) - but wasn't this one of the use-cases for Google's NotebookLM as well? Feedback: Many times when I'm reading a paper on arxiv - I find myself needing
825.
by vunderba 8mo ago
Pre-wikipedia days means you would likely have found me hunched over a 14" VGA monitor reading articles about giant devil rays and mansa musa in Microsoft Encarta as a kid. https://en.wikipedia.org/wiki/Encarta
826.
by vunderba 8mo ago
Yeah it's not perfect though I'll have to take your word for it - I looked up the concept of "bokeh swirl" but didn't really understand it in the context of the image. To me it looks like the two foreground subjects
827.
by vunderba 8mo ago
Thanks I appreciate it!
828.
by vunderba 8mo ago
You're totally right. THROW THE WHOLE THING OUT! Seriously thanks - will update the site.
829.
by vunderba 8mo ago
Results are in for `gemini-3.1-flash-image-preview` (NB 2) for the GenAI Showdown site in the editing comparisons. Remember to click the "Pass/Fail" button to toggle between pass/fail and a weighted score to account for
830.
by vunderba 8mo ago
Right? That part kind of blew my mind too - the multimodal model actually altered the overall lighting in the room by eliminating all the reflections and specular highlights when the natural light was taken away WITHOUT being asked in the
831.
by vunderba 8mo ago
This. Gpt-image-1/1.5 are the only ones that have this built in - though I'd love to have an insider view if its natively considering the alpha channel or just feeding it through a rembg-style post processor.
832.
by vunderba 8mo ago
NB Pro can do some seriously impressive edits around interior decorating - see the prompt that replaces the window with a mirror which correctly reflects the room. It's not perfect, but it's still damn impressive. https:/&#x
833.
by vunderba 8mo ago
> At this point I think identifying a work as AI-created makes people instantly devalue it. There was a study around this exact thing: https://mitsloan.mit.edu/ideas-made-to-matter/study-gauges-h...
834.
by vunderba 8mo ago
As somebody who juggles both figuratively and literally a lot of hobbies, I can definitely relate! One of my friends is a bit like you, they tend to experience the sudden flash of interest in various hobbies, dive extremely deeply and then
835.
by vunderba 8mo ago
> If I see a million fake Tom Cruise videos, then it oversaturates my desire for desire for all Tom Cruise movies. I often call this over-saturation the media equivalent of semantic satiation. Anything commoditized or mass-manufactured
836.
by vunderba 8mo ago
Jaded, but if I knew there was a possibility of a bunch of incriminating footage of me (images, video, etc.) out there in the pre-AI days, I would do my absolute best to flood the internet with as many related deepfakes (including of myself
837.
by vunderba 8mo ago
Agree just from a text search perspective alone that Mermaid even ASCII diagrams are usually preferable.
838.
by vunderba 8mo ago
I've only had a brief opportunity to try out NB Pro 2 (`gemini-3.1-flash-image-preview`), so I haven't had a chance to update GenAI Showdown. Here's some of my captions that tend to trip up even state-of-the-art models. http
839.
by vunderba 8mo ago
There's also an interesting set of benchmarks around giving LLMs a sort of "building-block harness and tools" to see how well they do putting together voxel-style models. https://github.com/Ammaar-Alam/mi
840.
by vunderba 8mo ago
Same. While it doesn't help that their name is about as generic as it gets, I searched across Kagi, Google, etc. and couldn't find any such YC company. That being said, it wouldn't entirely surprise me if somebody's trie

More ›