nmfisher

That was my point about coding agents - as long as there's a standard protocol (usually HTTP REST), Claude Opus could roll out a client library in almost the same amount of time it would take me to find the right name for the official pip package. Equally, they make it a lot easier for providers to roll out official SDKs. I did this recently for Tencent's Hunyuan API and I didn't even have to think about it. This type of API integration will be trivially solvable in the near future.

As a long time Dart/Flutter developer, I think Dart is slowly making its way to the server too. It’s more performant than Python (and I assume Ruby too), and nicer to work with than other statically typed languages (which I guess are mostly JVM or CLR based).

The third party package ecosystem is smaller but I think this will become less and less relevant as coding agents get better.

For what it’s worth, I unequivocally agree with you.

But if you spent any time with the older generation over the holiday break, it’s immediately obvious why it makes no difference. They sit there and scroll Facebook and Instagram advertising slop for multiple hours literally every single day.

Facebook will keep hoovering up cash until at least that generation dies off (and possibly the one after that too).

For what it's worth, I personally have a lot of respect for people who do this (or at the very least, people who forego higher salaries to avoid working for companies they find morally objectionable).

> they look at it and return it. The ID card doesn't get scanned

Actually in Australia, IDs usually do get scanned and stored. About the same time I was getting too old for clubs, they were starting to introduce ID scanners. You line up, hand over your driver's licence or passport, they slap it on a wall-mounted scanner, the scan goes into a database and in you go. No scan, no entry. Nowadays I think they just use phone/tablet scanners.

I went through the TS positive vetting process (for signals intelligence, not writing software for fighter jets, but the process is presumably the same).

If I were back on the job market, I’d be demanding a big premium to go through it again. It’s very intrusive, puts significant limitations on where you can go, and adds significant job uncertainty (since your job is now tied to your clearance).

I have an add-on that does the same thing as OP, with a free (open source) and paid version. The paid version lets you record direct to Blender.

https://nickfisher.gumroad.com/l/tvzndw

There are numerous criticisms you can level at Zuckerberg, but writing the first version of Facebook in PHP is not one of them.

That was how I took most of my school and university exams. I hated it then and I'd hate it now. For humanities, at least, it felt like a test of who could write the fastest (one which I fared well at, too, so it's not case of sour grapes).

I'd be much more in favour of oral examinations. Yes, they're more resource-intensive than grading written booklets, but it's not infeasible. Separately, I also hope it might go some way to lessening the attitude of "teaching to the test".

I didn't look too close but it wouldn't surprise me if this was intentional. Many of these Meta/Facebook projects don't have open licenses so they never graduate from web demos. Their voice cloning model was the same.

I took "kids" more to mean they're inexperienced at building sovereign data centres.

$ git clone repo && cd repo $ claude

Ask away. Best method I’ve found so far for this.

I wasn't actually able to use it because the servers were overloaded. What exactly impressed you (or more generally, what does it actually let you do at the moment?).

What was the 90 minute movie?

I'm sure Sam A will be fine, an IPO will probably see him ride off into the sunset with billions.

The average public investor buying pre-IPO shares, though, is a different story.

I don’t know the exact numbers, but I feel like OpenAI raised far more money than those companies, burned through it far quicker and has much more competition with a much shakier value proposition.

They definitely have a strong consumer brand so it’s not like they’re going to disappear, but I understand the bear case.

Yes, quantization isn't anything new, nor are audio codecs. As you point out, though, it's not just about designing a quantization scheme to reconstruct an analog signal. The scheme itself needs to be "easy" for current model architectures to learn and decode autoregressively (ideally, in realtime on standard hardware).

The blog post addresses this directly with samples from their own baseline (an autoregressive mu-law vocoder), and from WaveNet (which was similar architecture). The sound is mostly recognizable as a human voice, but it's unintelligible. The sequence length is too long and the SNR for the encoding scheme is too low for an generative/autoregressive model to learn.

This is what the neural codec is intended to address. Decoupling semantic from acoustic modelling is an important step ("how our ears interpret a sound" vs. "what we need to reconstruct the exact acoustic signal"). Mimi works at 1.1kbps, and others work at low bitrates (descript, semanticodec, etc). Encodec runs at at a higher bitrate so generally delivers better audio quality.

Now - why are neural codecs easier to model than conventional parametric codecs? I don't know. Maybe they're not, maybe it's just an artifact of the transformer architecture (since semantic tokens are generally extracted from self-supervised models like WavLM). It's definitely an interesting question.

The article is talking about doing exactly that. The key question is how to convert an inherently continuous signal (speech/audio) into a discrete set of tokens. A single window of audio is usually somewhere between 10ms and 100ms. It's difficult to squish all that information down to a single "token" that represents the semantic and acoustic content for that window.

That's why residual vector quantization is a useful technique - using multiple dictionaries to quantize a single timeslice, each conditioned on the previous residual level. You can also quantize a signal at different frequencies.

There are samples towards the end of the post of their LLM trained on their Mimi audio codec.

Interesting - do you have the source handy?

Overall I'm quite positive towards core Cloudflare products like Tunnels, Workers, R2, KV etc, but a lot of newer products are often either thoroughly broken (e.g. Cloudflare AI) or unusable due to insufficient documentation (e.g. Email Routing).

After being burned a few times, I think I'm going to ignore any new Cloudflare product for 12 months after stable release. If their products worked as advertised, I'd be willing to pay considerably more. I think their commitment to the free tier is hamstringing them a little bit.

Ubank in Australia just told me they’re retiring their website in a few months, the app will be the only way to access your account. It’s digital only, so no real world branches either.

Feels very much like a knee-jerk response to Facebook releasing their "Vibes" app the week before. It's basically the same thing, OpenAI are probably willing to light a pile of money on fire to take the wind out of their sails.

I also don't think the "Sam Altman" videos were authentic/organic at all, smells much more like a coordinated astroturfing campaign.

Dart transpiles to Javascript already - not exactly native support, but practically the next best thing.

That being said, I'm also 100% behind the effort to standardize WASM as the cross-platform compilation target of choice.

Why do you say they are different models? I've been looking at this today and haven't seen anything explicitly state that.

A dollar is always a dollar, so it's hard to claim that $1 million in revenue is actually worth $10 million. OpenAI shares, on the other hand, aren't publicly traded, so it's much easier to claim they're worth $10 million when noone would actually be willing to buy for more than $1 million.

It's not necessarily manipulative but it's also not exactly an arms-length purchase of GPUs on the open market.

> We need to build better native UI libraries that just open up a WebGL context and draw shit to that.

This is what Flutter does. It works well, but you do lose some nice things that the browser/DOM provides (accessibility, zooming, text sizing/selection, etc). There’s also a noticeable overhead when the app + renderer is initially downloaded).

Singapore definitely does have this problem, there are police/government warnings against scams and fraud plastered everywhere.

Were there many competitors to YouTube though? I remember Vimeo (still around) and Google Video (replaced by YouTube), but not much else.

Between OpenAI, Anthropic, Google, Facebook, xai, Microsoft, Mistral, Alibaba, DeepSeek, z.ai, Falcon, and many others, AI feels a lot more competitive.

Filament is a great lightweight cross-platform PBR rendering library. There's a few things missing (real GPU instancing is one), but in terms of shadows/antialiasing/color grading, it's very powerful. The material (shader) language is also very accessible.

I maintain an open-source Dart/Flutter package[0] which is mostly a wrapper around Filament. This makes it considerably easier to have a single UI+codebase that runs across macOS, iOS, Android, Windows, and Web.

[0] https://github.com/nmfisher/thermion

Didn't they almost go broke after the Wii U? I vaguely remember that the Switch was do-or-die for them.