Hacker News: santiagobasulto

New comment by santiagobasulto in "Gemma 4 12B: A unified, encoder-free multimodal model"

santiagobasulto — Wed, 03 Jun 2026 19:29:50 +0000

Not at all, I had the same feeling as yours the first time I read it. I think the key is that the "encoder" they're using is just a linear projection, which is probably pretty fast and memory efficient. A single matmul vs a ViT encoder is probably a huge win.

New comment by santiagobasulto in "Gemma 4 12B: A unified, encoder-free multimodal model"

santiagobasulto — Wed, 03 Jun 2026 19:27:45 +0000

I don't think it's the same. It's a similar concept, but Gemma is using just a linear projection, which I assume is a lot faster. The developer guide has more details: https://developers.googleblog.com/gemma-4-12b-the-developer-...

    Vision embedder (35M parameters): Replaces the 27 vision transformer layers of the other medium-sized Gemma 4 models. Raw 48x48 pixel patches are projected to the LLM hidden dimension with a single matmul. A factorized coordinate lookup (X and Y matrices) attaches spatial location information directly to the input

the "single matmul" is the key here, I haven't tried it, but it's probably pretty fast and memory efficient.

New comment by santiagobasulto in "The $500K AI Film That "Premiered at Cannes" Was Not in the Official Festival"

santiagobasulto — Fri, 29 May 2026 11:52:03 +0000

I mean, "Cannes" the city right? lol

New comment by santiagobasulto in "Ask HN: Is anyone working at least 4 hours daily on an Apple Vision Pro?"

santiagobasulto — Tue, 26 May 2026 06:39:02 +0000

I don't own Apple Vision Pro, but I tested them two times already and I'm amazed by the technology. I honestly think it's the future. We're probably WAY too early yet, and the Vision Pro might fail badly, but it is indeed the future.

I did a special test session in Japan for "productivity" (the guys at the Apple Store were very friendly and agreed to let me install VSCode and Ghostty on the testing laptop. I cloned an open source repository and spent ~20 minutes just coding.

It was FANTASTIC. The Apple Store was full and I could still "black out" the noise and completely immersed myself in the experience.

I'm seriously considering buying a pair now, but I'm just concerned about the under-investment in the sector.

Regardless, I honestly think it's the future, maybe in 10/20 years, but it'll be the norm.

New comment by santiagobasulto in "Ask HN: How did you land your first projects as a solo engineer/consultant?"

santiagobasulto — Sun, 19 Apr 2026 10:27:03 +0000

General consultancy is an extremely crowded space. As a startup CEO, I get at least 3 emails per week from software agencies and consultants. On top of that, they're usually located in India/Ukraine and the rates they offer are very low, so I assume it's very difficult to compete.

My advice would be to differentiate yourself:

- Become an expert in 1 thing, and one thing only: either start an open source project, or become the main collaborator in one. And be an EXPERT in that ONE thing. Not a generalist.

- Go personal: I can't see who you are or where are you based in your website. If I want to hire an EXPERT (see point before) consultant, I want to see their face and why they're different. I need a feeling of trust.

- Network the hell out of it: once you're an expert on one thing and you have a face, people will recognize you and recommend you

New comment by santiagobasulto in "Muse Spark: Scaling Towards Personal Superintelligence"

santiagobasulto — Wed, 08 Apr 2026 16:41:15 +0000

This looks like a very interesting model and very promising, especially after llama lost so much ground recently. I hope they release the weights

New comment by santiagobasulto in "Why Sora Failed: $15M/day inference cost vs. $2.1M lifetime revenue"

santiagobasulto — Thu, 26 Mar 2026 10:01:30 +0000

But doesn't $15M/day of inference cost imply "demand" from users? If this is the case, it's just a matter of time until costs can be reduced.

New comment by santiagobasulto in "Tell HN: Litellm 1.82.7 and 1.82.8 on PyPI are compromised"

santiagobasulto — Tue, 24 Mar 2026 14:33:05 +0000

I blogged about this last year[0]...

> ### Software Supply Chain is a Pain in the A*

> On top of that, the room for vulnerabilities and supply chain attacks has increased dramatically

AI Is not about fancy models, is about plain old Software Engineering. I strongly advised our team of "not-so-senior" devs to not use LiteLLM or LangChain or anything like that and just stick to `requests.post('...')".

[0] https://sb.thoughts.ar/posts/2025/12/03/ai-is-all-about-soft...

New comment by santiagobasulto in "Fyn: An uv fork with new features, bug fixes, stripped telemetry"

santiagobasulto — Mon, 23 Mar 2026 13:58:31 +0000

Sometimes I want the venvs to be in a centralized location, and just do:

UV_PROJECT_ENVIRONMENT=$HOME/.virtualenvs/{env-name} uv {command}

New comment by santiagobasulto in "Give Django your time and money, not your tokens"

santiagobasulto — Tue, 17 Mar 2026 20:52:04 +0000

Yes. Exactly. We’re both “feeling” without much proof. But between the two speculations, one is more open and welcoming, while the other is more restrictive.

New comment by santiagobasulto in "Give Django your time and money, not your tokens"

santiagobasulto — Tue, 17 Mar 2026 16:17:35 +0000

I feel like open source is taking the wrong stance here. There’s a lot of gatekeeping, first. And second, this approach is like trying to stop a tsunami with an umbrella. AI is here to stay. We can’t stop it, for much we try.

I feel the successful OS projects will be the ones embracing the change, not stopping it. For example, automating code reviews with AI.

New comment by santiagobasulto in "AirPods Max 2"

santiagobasulto — Mon, 16 Mar 2026 19:14:36 +0000

I’ve been using my AirPod Max for hours for the past 2 years and never noticed they were “heavy”. I’m wondering now as I’ve never researched on headphones (I just buy simplicity from Apple, I’m not an audio sophisticated costumer) that was never brought out to me, so I haven’t even noticed.

New comment by santiagobasulto in "Global warming has accelerated significantly"

santiagobasulto — Fri, 06 Mar 2026 16:14:59 +0000

What! you're saying that my selective recycling of paper and having the plastic caps attached to the bottles didn't work? SHOCKING

(only europeans will understand)

New comment by santiagobasulto in "How to code Claude Code in 200 lines of code"

santiagobasulto — Fri, 09 Jan 2026 12:35:39 +0000

I'm surprised this post has so many upvotes. This is a gross oversimplification of what Claude Code (and other agents can do). On top of that, it's very poorly engineered.

New comment by santiagobasulto in "Rust is beyond object-oriented, part 3: Inheritance (2023)"

santiagobasulto — Wed, 07 Jan 2026 09:06:07 +0000

Yes I agree with you. The pattern is "composition" vs "inheritance". Defining a "thing" as "what it can do" instead of "what it is". Instead of saying that "a duck is a Bird which in turn is an Animal which in turn is a LivingThing" (Duck -> Bird -> Animal -> LivingThing) you focus on what a duck can do: a duck "quacks, swims, etc":

    class Duck(Swimmable, Quackable, FishEatable...)

I think there's still a place for "inheritance" based approach for APIs that need to be very strict about subtyping: would be hard to express covariance/invariance/contravariance without it.

New comment by santiagobasulto in "Databases in 2025: A Year in Review"

santiagobasulto — Mon, 05 Jan 2026 09:27:49 +0000

I love these yearly review posts. Thanks Andy and team.

New comment by santiagobasulto in "Ask HN: Who is hiring? (January 2026)"

santiagobasulto — Sat, 03 Jan 2026 21:37:24 +0000

I’m on chrome. Maybe it was a hiccup on my end.

Looks interesting.

We are in the process of looking for some alternatives to temporal/prefect. Would you mind sharing your email so I can send a few questions along with my cofounder?

New comment by santiagobasulto in "Ask HN: Who is hiring? (January 2026)"

santiagobasulto — Sat, 03 Jan 2026 11:45:29 +0000

Your website is INCREDIBLY slow, and I have a very good laptop with dedicated GPU. Somehow some JS is killing the whole thing.

Anyways, the concept looks cool, but I'm failing to see a real value add to something like Temporal. What is it?

New comment by santiagobasulto in "Show HN: Stop AI scrapers from hammering your self-hosted blog (using porn)"

santiagobasulto — Fri, 19 Dec 2025 12:40:22 +0000

Offtopic: when did js/ts apps get so complicated? I tried to browse the repo and there are so many configuration files and directories for such a simple functionality that should be 1 or 2 modules. It reminds me of the old Java days.

New comment by santiagobasulto in "The History of Xerox"

santiagobasulto — Mon, 15 Dec 2025 09:03:46 +0000

Very interesting story and well written so far, I'll finish it after work.

One very interesting thing about Xerox was not only their technology but their choice of business model. As smaller companies couldn't afford an expensive copier, they'd "rent" it and charge per copy. From the article:

> The company placed machines in well-traveled public spaces where it was on display, and in addition to sales, they also offered machine rental for smaller organizations. This was a low price for up to 2000 copies, and each copy after was 4¢. They also promised that a machine could be returned within fifteen days. The 650 pound behemoth was wildly successful.

Another similar interesting business model was pioneered by Rolls-Royce in their airplane turbine business. Instead of selling their whole turbine, they'd "rent" it and charge it "per flight hour", derisking both parts.