dannyw

16 ms·

421.
by dannyw 5mo ago
Building distributed training pipelines or optimising your ML stack (examples called out in the model card) isn’t harmful.
422.
by dannyw 5mo ago
What kind of work are you getting refusals on? Genuinely curious. The only refusal I’ve had in recent memory was declining to find doorbell camera footage matching a certain description, which is fair enough and I think EU laws heavily rest
423.
by dannyw 5mo ago
The trouble is the silence, not Anthropic setting guardrails. Claude saying "I'm sorry, I can't assist further because it looks like you're [XYZ]" is fine. We all know the false positive rates for classifiers on Fab
424.
by dannyw 5mo ago
Competition is lovely. And ironically, OpenAI will probably get and keep lots of enterprise customers like Microsoft^ that won’t accept anything less of ZDR. [1] https://www.theverge.com/report/947575/microsoft-cla
425.
by dannyw 5mo ago
Sorry, I’m specifically referring to the silent degradation of the model to “limit frontier LLM development”. From the description, it appears to encapsulate far more than frontier LLM development, but general ML research and development to
426.
by dannyw 5mo ago
Agreed. Distillation is as much of an attack as scraping is an attack ;)
427.
by dannyw 5mo ago
Fable is really, really good at this. My workflow involves giving it a bit of human inspiration, through asking it to generate a few different design templates/scaffolds (without building the whole frontend first). Then I iterative and
428.
by dannyw 5mo ago
I wouldn’t over interpret this. It could be a bad summary of “what solution would be more valuable to the user?”
429.
by dannyw 5mo ago
There’s something really fun about writing your own microcontroller code as a software engineer that hasn’t worked with embedded before. At least for me. Before you just vibe code, consider if it piques your interest. You might just enjoy l
430.
by dannyw 5mo ago
That I can understand. It’s Anthropic’s right to choose their customers. But silent degradation for use cases including “distributed training” as one of their examples is going to catch up a lot of proper use cases. Not everyone in AI or ML
431.
by dannyw 5mo ago
They have all transcripts for at least 30 days. The problem is that (as anyone who used Fable can attest) their classifiers are extremely sensitive and catch tons of innocent queries. Imagine being a data scientist or MLE training a small c
432.
by dannyw 5mo ago
No, that’s for “frontier LLM development” which somehow includes examples like distributed training infra. Based on how sensitive the classifers are, any data scientist / MLE is probably going to encounter cases where some silent degra
433.
by dannyw 5mo ago
The challenge is the examples they’ve mentioned (distributed training infra? ML acceleration techniques?) go beyond what’s prohibited by their ToS and is like a catch net. I would wager the majority of ML and data science work in the world
434.
by dannyw 5mo ago
It’s not binary. With AWS previously you have contractual guarantees with a third party, that’s been in business for a couple decades, which explicitly state zero seconds of data retention - only as long as needed for inference. Consider th
435.
by dannyw 5mo ago
Yes, that is your intended purpose of “git push”, it’s to save. And only if you use GitHub. A better analogy here is probably “every time you use VS Code, the files you edit get sent to Microsoft”. Some legitimate concerns: • You have trade
436.
by dannyw 5mo ago
When there is a precise and legally defined boundary (i.e. ZDR means your data with Bedrock stays within the Amazon security and legal boundary), it becomes significantly more difficult to hide full data egress; without alarm bells being ra
437.
by dannyw 5mo ago
Yep. We have some interesting problems, like getting LLMs to create/edit Canva designs in our own proprietary format, which isn’t published or documented on the web. So the model has to work with it, purely from a very detailed system
438.
by dannyw 5mo ago
Impressions from testing Fable 5 prior to launch: • My most noticeable immediate jump was in how its frontend design was much more intentionally crafted, and delightful without feeling like 'AI vibe coded'; with better end-user us
439.
by dannyw 5mo ago
Can I ask what your task and application is? A ~20% failure rate sounds atypical. If you’re slightly hyperbolic and mean something like 2-5%, yeah that’s a property of LLMs; but also heavily affected by how you prompt and how you constrain
440.
by dannyw 5mo ago
Not for API pricing. The latest models are not subsidised API wise anymore. Qwen3.6 is practically indistinguishable to Sonnet 4.6 at least in my personal experience. And sonnet 4.6 is not that cheap.
441.
by dannyw 5mo ago
No big tech company hands write code by default now. You’d get PIP’d very quickly when your manager notices your PRs don’t have AI as a co-committer.
442.
by dannyw 5mo ago
These companies are allowed to go public and anyone can buy their shares. Since the start, the S&P 500 has had a simple and consistent profitability screen. Your company must be GAAP profitable in the past quarter, as well as for your p
443.
by dannyw 5mo ago
Swapping to a NVMe will also consume PE cycles on your NAND, ie wearing it out over time. RAM/VRAM don’t degrade from use.
444.
by dannyw 5mo ago
Yeah, no production or large scale VC system is running software AV1 encoders on smartphones. You will drain a full phone battery in 1-2 hours of calls. It just doesn’t make sense and will result in extraordinary power/battery drainage
445.
by dannyw 5mo ago
There's a huge difference between everything you mentioned, and dealing with 'browser quirks' like: <!DOCTYPE html> <!--[if IE 8]> <html lang="en" class="ie8 no-js"> <![endif]--&
446.
by dannyw 5mo ago
I used to make a living doing frontend development, and quirks and knowing idiosyncrasies is a burden to your craft. Yes, it meant there were higher barriers to entry, but it also resulted in a lot of broken websites, and I can tell you it
447.
by dannyw 5mo ago
I still use 4.7. I don’t know what I’m doing wrong but 4.7 frequently tells me to it’s time to sleep at all hours of the day while working. I’ve tried clearing all my memory/agents files. I’m hoping the “go to sleep” behavior has been
448.
by dannyw 5mo ago
It is not a trivial challenge setting up model serving infra for ~1T or larger models, especially in a high reliability environment (e.g. your team is using it for work, or you're using it to power production apps). Sure, there are thi
449.
by dannyw 5mo ago
Honestly LineageOS is probably a more secure root than the typical random android OEM; unless you're dealing with Samsung or Google.
450.
by dannyw 5mo ago
You can now. You couldn't do this in the early versions of iOS.

More ›