arnaudsm

8 ms·

61.
by arnaudsm 1y ago
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.
62.
by arnaudsm 1y ago
Basic visibility should be the default, most people never change their settings.
63.
by arnaudsm 1y ago
On top of wasting GPU cycles, such low-contrast graphics are terrible for older users. The Apple Music navbar is hilariously unreadable and distracting.
64.
by arnaudsm 1y ago
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)
65.
by arnaudsm 1y ago
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.
66.
by arnaudsm 1y ago
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.
67.
by arnaudsm 1y ago
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 This post will be outdated
68.
by arnaudsm 1y ago
That's a lot of kilo-watt-hours wasted for a find and replace operation. Have you heard of text.replace("—", "-") ?
69.
by arnaudsm 1y ago
New model homepage : https://deepmind.google/technologies/gemini/ Old model card : https://storage.googleapis.com/model-cards/documents/gemini-... They intentionally buried that informat
70.
by arnaudsm 1y ago
From a business pov it's a great move, for the customers it's evil to hide evidence that your product became worse.
71.
by arnaudsm 1y ago
They should have shown the benchmarks. Or market it as a coding model, like Qwen & Mistral.
72.
by arnaudsm 1y ago
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.
73.
by arnaudsm 1y ago
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.
74.
by arnaudsm 1y ago
thanks, early 30s are exhausting haha
75.
by arnaudsm 1y ago
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
76.
by arnaudsm 1y ago
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)
77.
by arnaudsm 1y ago
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,
78.
by arnaudsm 1y ago
It's probably a combination of both innovation and quality. Respecting your users matters.
79.
by arnaudsm 1y ago
It's the pressure of the investors that leads to enshittification. Bootstraped and family businesses are less likely to do it.
80.
by arnaudsm 1y ago
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.
81.
by arnaudsm 1y ago
I agree, the monopolistic practices are concerning, especially against the open internet.
82.
by arnaudsm 1y ago
That's a scary scenario but you may be right
83.
by arnaudsm 1y ago
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.
84.
by arnaudsm 1y ago
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.
85.
by arnaudsm 1y ago
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.
86.
by arnaudsm 1y ago
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 entrepr
87.
by arnaudsm 1y ago
I partly disagree, software still sucks and it's a great time to build. The #1 OS is slow and crashes all the time. The #1 email client takes 10s to load on my mother's laptop. Most popular products are slow, buggy, filled with sp
88.
by arnaudsm 1y ago
I get often ratelimited or blocked from websites because I browse them too fast with my keyboard and mouse. The AI would be slowed down significantly. LLM-desktop interfaces make great demos, but they are too slow to be usable in practice.
89.
by arnaudsm 1y ago
The complexity of the web specs is out of control. It is a form of regulatory capture from Google & Apple to prevent new players from entering the space. Just look at the devellopment hell of Servo & Ladybird. Maybe the future is in
90.
by arnaudsm 1y ago
I've witnessed projects beind rewritten using Cursor, with 10x more code & dozens of bugs no one noticed. Vibe coding is the new offshoring.

More ›