arnaudsm

This is great, there are too few open-source video editing projects!

However, why do you have a waitlist for an open-source project? And why do you brag of having Databuddy analytics built-in?

Are you just a startup that's building in public and using open-source as a marketing trick before monetizing it?

Remembering the failure of Google+, I wonder if hostilely forcing a product to your users makes it less likely to succeed.

If they become self improving, the first one would outpace all the other AI labs and capture all the economical value.

- https://livebench.ai/#/ + AIME + LiveCodeBench for reasoning

- MMLU-Pro for knowledge

- https://lmarena.ai/leaderboard for user preference

We only got Magistral's GPQA, AIME & livecodebench so far.

I wished the charts included Qwen3, the current SOTA in reasoning.

Qwen3-4B almost beats Magistral-22B on the 4 available benchmarks, and Qwen3-30B-A3B is miles ahead.

Basic visibility should be the default, most people never change their settings.

On top of wasting GPU cycles, such low-contrast graphics are terrible for older users. The Apple Music navbar is hilariously unreadable and distracting.

No, there's the original voice mode (4o+tts), the original AVM (4o with native audio-to-audio), the new AVM announced in this post, and the mini AVM (4o-mini with native audio-to-audio)

If there's an OpenAI PM reading this: please add the model selector for voice modes. 80% of this thread is users confused about which model they're using.

We are about to enter the era of aggressive LLM monetisation and anti-competitivness. We got used to cheap subsidized models, in bet in 2 years we'll pay double for the same service.

Qwen3 family (and the R1 qwen3-8b distill) is #1 in programming and reasoning.

However it's heavily censored on political topics because of its Chinese origin. For world knowledge, I'd recommend Gemma3.

This post will be outdated in a month. Check https://livebench.ai and https://aider.chat/docs/leaderboards/ for up to date benchmarks

That's a lot of kilo-watt-hours wasted for a find and replace operation.

Have you heard of text.replace("—", "-") ?

New model homepage : https://deepmind.google/technologies/gemini/

Old model card : https://storage.googleapis.com/model-cards/documents/gemini-...

They intentionally buried that information

From a business pov it's a great move, for the customers it's evil to hide evidence that your product became worse.

They should have shown the benchmarks. Or market it as a coding model, like Qwen & Mistral.

Be careful, this model is worse than 03-25 in 10 of the 12 benchmarks (!)

I bet they kept training on coding, made everything worse on the way, and tried to hide it under the rug because of the sunk costs.

This should be the top comment. Cherry-picking is hurting this industry.

I bet they kept training on coding tasks, made everything worse on the way, and tried to hide it under the rug because of the sunk costs.

thanks, early 30s are exhausting haha

The biggest impact Generative Music had on my life was for wedding skits. Families that wanted a funny song about the bride filled with anecdotes didn't need talent anymore.

The first time I heard it, it was incredible. The 2nd wedding that did it, it started to feel boring. The 3rd time, everyone hated it.

Similar to image-generation, we're getting tired really fast of cookie-cutter art. I don't know how to feel about it.

Production budget != Compute budget. Minecraft, a 15 year old game, runs at 30fps on the Quest2, which will make anyone instantly sick.

Although I've had a blast on Minecraft in PCVR (but you need strong VR legs)

400h+ VR gamer too, I disagree on "no crazy graphics". The Standalone strategy of leaving PCVR made VR accessible, but killed game innovation entirely. The entire industry was downgraded graphically of 2 entire generations. Sure, there are wizards like StressLevelZero that can ship beautiful games on Quest, but most studios cannot.

I wish they hadn't abandonned PCVR, and kept pushing the boundaries of the medium like Valve did with Alyx.

It's probably a combination of both innovation and quality. Respecting your users matters.

It's the pressure of the investors that leads to enshittification. Bootstraped and family businesses are less likely to do it.

You're correctly describing the most common mindset in tech entrepreneurs. Chasing the last shiny trend instead of caring about quality.

Apple is the best example that doing the opposite can be immensely successful.

I agree, the monopolistic practices are concerning, especially against the open internet.

That's a scary scenario but you may be right

Schumpeter's cycle of destructive creation usually takes 15 years in software.

IBM, Microsoft, Apple, Google, Facebook. They were all the cool underdogs for a decade, then slowly decline in quality.

Good question, AI coding is currently limited by how fast you can review it. I still see Gemini 2.5 Pro output insane vulnerabilities once in a while.

You may get a 2x speedup, but humans are still the bottleneck.

I meant popularity. My brand new Win11 laptop freezes or crashes all the time. My relatives laptops too. It's slower every year, even on high-end hardware.

Users have inertia, it takes years for superior products to win.

VC capital is useful for speed & innovation, but most of the time leads to bloat & rent-seeking. Does DocuSign need 7,000 employees? My journey of bootstrapped entrepreneurship has been much more sustainable and respectful to my users.