dannyw

10 ms·

121.
by dannyw 2mo ago
We have a simple definition for this: COGS. We also have another solution for "whatever accounting decides": generally accepted accounting practices. It's far from perfect, but GAAP figures are what you should be looking at;
122.
by dannyw 2mo ago
A surprising amount of companies sell exact the same product through different channels for different prices. A good example is Apple. Multiple times a year, retailers here in Australia have co-ordinated sales on Apple products. Apple.com o
123.
by dannyw 2mo ago
That suggests this is being funded by OpenRouter, and there's no indication this is the case (and I doubt OpenRouter can afford it; why would they anyway).
124.
by dannyw 2mo ago
Yes, but their shapes are very different. Most users interact with a small, geographically constrained bubble. It is (comparably) an easier problem for the usage patterns of Facebook. The app is also designed so when you open Insta, only a
125.
by dannyw 2mo ago
You can train a LLM to inverse summarised thinking into thinking text. It’s not perfect, but it gets you maybe 80% of the quality with proper techniques. Paper: https://arxiv.org/abs/2603.07267 FWIW, there’s not that m
126.
by dannyw 2mo ago
The network effects isn’t that much, and like what we see with Steam, you want to be where the audience is. More realistically, MAU or something is a metric/OKR. It’s easy to imagine why a business wouldn’t want to cap that.
127.
by dannyw 2mo ago
I’d love to see the org chart sizes How many hires went into infra and platform? How many went to Gen AI org units? From what I see, and that is a constant wave of very meh features, it seems like the latter.
128.
by dannyw 2mo ago
Services like Facebook aggressively block automated activity, including AI agents. GitHub sells this as a service.
129.
by dannyw 2mo ago
Or maybe prioritise humans / human-shaped activity? Agents can tolerate a 503+Retry-After is a lot less frustrating to an agent than a human.
130.
by dannyw 2mo ago
More likely, as seen in a lot of companies, perhaps infra teams got whittled down (or frozen HC, less than BAU, etc) with resources reallocated to AI org units.
131.
by dannyw 2mo ago
Where would you put comments then? Sure, there’s always mailing lists, but even so, someone still has to host Mailman. I do think GitHub has positively contributed to the software industry and open source. It’s been enshittified for sure, b
132.
by dannyw 2mo ago
I find the ‘explanatory’ output style of Claude to be a bit more tolerable, but yes. Claude seems to speak and write more in Claude-speak with every release.
133.
by dannyw 2mo ago
Your configuration is broken or wrong. What are you using? Hopefully not llama.cpp? I’ve sweeped concurrency across many models and many different kinds of hardware, and the only times I saw similar results to you were when I didn’t configu
134.
by dannyw 2mo ago
Gemini is honestly an excellent LLM with many capability strengths. For example, 3.7 Flash is #1 on MMLU Pro and AA’s agentic spreadsheets/docs benchmark, etc. Yes, beating Fable. Agentic coding is only one dimension.
135.
by dannyw 2mo ago
Open weight models did take off significantly over the past couple of months. You can see their exponential curve in https://openrouter.ai/rankings#top-models It's so significant that even OpenAI is doing promos on Ope
136.
by dannyw 2mo ago
The model itself is excellent, the defaults are bad. As Simon and other pointed out, medium is great. Reminds me of Gemma4 and the official (or at least popularly used around launch) Jinja templates being wrong and broken for tool calling.
137.
by dannyw 2mo ago
The recent 'Stolen Thoughts'[1] paper shows many excerpts of private reasoning for frontier models. For a complex maths problem, Sol reasoned in 367 tokens before working: We need solve. Need interpret no person sits next to t
138.
by dannyw 2mo ago
There’s a lot of different new usernames with those non-constructive comments, but I wouldn’t be surprised if it’s just a couple very persistent people making new accounts after given a suspension or something.
139.
by dannyw 2mo ago
If you're a company with a considerable API bill, buying a few Blackwells and a rack could make a lot of economic sense, while still giving your security team full control of the infrastructure
140.
by dannyw 2mo ago
Models are trained in highly distributed and parallel ways these days. It is entirely possible to run multiple concurrent post-training runs. When a frontier lap deploys a 1M RL gym rollout, these 1M environments are absolutely not talking
141.
by dannyw 2mo ago
Have you counted your thinking tokens for say Opus or Fable? It wouldn’t surprise me if frontier closed models “over-reason” just as much, but you don’t see it thanks to the summariser. (We do know GPT5.6 have adopted the caveman shorthand,
142.
by dannyw 2mo ago
The link doesn’t seem to have any info about thinking behavior and patterns, or examples? It’s also difficult to trust summarised thinking from closed models. As we saw with GPT’s caveman, what and how it thinks about isn’t the friendly fir
143.
by dannyw 2mo ago
The official docs recommend different sampler options for thinking off. Have you tried adjusting those?
144.
by dannyw 2mo ago
It’s unlikely they are benchmarking that downstream. They probably have private benchmark scripts with purpose-specific run telemetry and logging, etc. I’m sure there is some basic testing but it might be agentic (LLM likes its own output)
145.
by dannyw 2mo ago
Trading blows in some benchmarks is a bit exaggerated. If you try the model, `xhigh` is basically feels like the `max` mode (i.e. massive thinker and extremely presistent), and the amount of world knowledge and intent understanding is nowhe
146.
by dannyw 2mo ago
`/output_style new` can be used to tailor the output style to your preferences, e.g. make it more literal and task-execution focused, and feel free to return control to the user when unclear; can keep work going in the background. The
147.
by dannyw 2mo ago
It's very agentic coding focused; and I'd say a good executor but certainly not Opus in scale; overall knowledge; long-horizon work and recovery; etc. e.g. If you try to chat to it about something philosophical for example, or may
148.
by dannyw 2mo ago
Probably helps it score a little bit better in benchmarks :) `medium` seems like a nice balance so far; along with some light steering to vary think effort as needed for task and being pragmatic.
149.
by dannyw 2mo ago
For comparison tho, 400GB of cloud storage for $5/mo is actually quite amazing even if it didn’t include anything else. I know, it’s cloud storage, a NAS lets you own your data, but for non techies who have a bunch of photos and videos
150.
by dannyw 2mo ago
I like to think of Gemini as a broader generalist that hasn't allocated _most_ of its skillpoints to agentic coding execution :)

More ›