HNHacker News
TopNewBestAskShowJobs

sroussey

5,695 karma · joined December 20, 2009

workglow.dev

Previously: Embarc, Privicy, Wag, Weebly, Firebug, Network54, Kantara

submissionscomments
sroussey··on GPT-5.6 Luna vs. GPT-6 Astra: Is a $1.20 Model Good Enough for Code Review?
Should this not all happen before the PR is created?
sroussey··on The case against JPEG XL
All my photos are HEIC. Why not just use what I have and not translate?
sroussey··on Flock cameras used to arrest a child for playing on a swing
How is your location data protected? They may not be able to force a company to hand it over without a warrant… but they can just buy it.
sroussey··on I'm being cyberattacked by Tesla, Inc
Give Tesla the wrong time.
sroussey··on CUDA for AMD on Windows
What models can it run?
sroussey··on CUDA for AMD on Windows
Hugging face is working on something like this where well known models get fused into a single implementation.
sroussey··on Astra and Fable still hack on simple variants of alignment evals from 2025
Language models have always had an issue with negatives.

A negative like do “not” xyz is just not encoded the same as spelling out what you want vs what you don’t want.

Harder to write though.

sroussey··on Waymo pulls over, calls cops on juvenile riders who had 'ghost gun"
> Private surveillance that immediately sends any and all crimes to law enforcement is indistinguishable from government surveillance and should treated the same way.

Don’t use Snapchat then…

https://www.gadgetreview.com/student-teacher-sent-a-private-...

sroussey··on Retrospectively Reverse-Engineering Apple's Neural Engine
Dictation, text to speech, some of that computational photography, etc too!
sroussey··on OpenAI Agents API
> All of this is specified in the ACP spec

Oh good reminder. I need to do that.

> Also vital is proper prompt caching, tool design and some connection retry mechanism.

prompt caching is weirder than i originally thought, and so variable across providers. Retry is easy, but can you explain more on tool design?

sroussey··on DeepSeek v4.1 Flash
People read the AWS rate card for bandwith and think has something to do with reality. Even though it is 1000x higher!
sroussey··on iPhone Duo
For me, I can never reach the bottom or top. Sideways has never been an issue.

I would like a non-foldable version of this phone. Call it the iPhone Wide or IPhone Mini Wide or whatever.

sroussey··on Apple Unveils iPhone Duo
The pro max is like 6.9” screen and the duo is 5.4”? How is that bigger? Perhaps wider, but the thumb should have better reach and coverage of the screen.
sroussey··on iPhone Duo
Right handed women (which is a giant market) are going to love this thing.
sroussey··on Apple Unveils iPhone Duo
You are describing me as well, but our market size is negligible compared to to 50% of the population that throws this thing in a purse.
sroussey··on Apple Unveils iPhone Duo
Are you left handed? The rearranging of the UI to provide navigation on the right side of the 5.4” screen seems like it was made for this exact purpose.
sroussey··on Paramount Caught Using 'Astroturf' Group to Drum Up Fake Support for Merger
The cost savings from consolidating things like HR, and the debt payments are different by orders of magnitude. The cost synergies will be a rounding error. Structural changes like stopping investment in the future, mass layoffs, and selling off assets (or rights to the assets) will be required.
sroussey··on LG TVs caught spying even when offline or on standby
All smart TVs should be disconnected from the internet. Use an AppleTV or similar.
sroussey··on Benchmarking Qwen3.8 27B quantizations: 4-bit holds up, 1-bit collapses
Have you tried https://huggingface.co/prism-ml/Bonsai-27B-mlx-1bit ? PrismML is the only people i am aware of doing 1bit that is decent.
sroussey··on US Military disables ad trackers on troops' phones
Yes, it was temporary and Meta revenue and profit at all time high (before data center buildout).
sroussey··on Ask HN: Why were OpenAI, Claude, and Grok simultaneously down?
Nope. I am getting Gemini errors now...
sroussey··on Ask HN: Why are OpenAI, Claude, and Grok simultaneously down? Coincidence?
I did, for stuff i do in cursor.

i also finally installed opencode and switched its model to muse 1.3

both are decent.

sroussey··on Claude Fable 5.1 and Claude Mythos 5.1
标
sroussey··on Claude Fable 5.1 and Claude Mythos 5.1
I can tell as it uses what are super accurate words but i only know that after i lookup the phase.

Today, i was stumped as it substituted a work with a word from another language!

"The four items the ledger had标 standing open..."

sroussey··on Claude Fable 5.1 and Claude Mythos 5.1
Or when it tries a different language to explain a word: "The four items the ledger had标 standing open"!
sroussey··on Claude Fable 5.1 and Claude Mythos 5.1
Anthropic in particular has been moving the level of watermarking since the beginning of the year. You didn’t think it just went from off to on one day did you? It needs time to get plastered over the internet, see what people do that break it, how Google summaries erase it (or not, or add their own). All of this takes at least months if not a year.
sroussey··on Muse Spark 1.3
You should post your source code you wrote here… ;)
sroussey··on WebLLM: high-performance in-browser LLM inference engine
And ONNX is what Transformers.js uses as well, at least for the moment.
sroussey··on Claude Fable 5.1 and Claude Mythos 5.1
No, I mean that when constrained (like discussing a specific bug in code) the options are limited.

SynthID specifically mentions something similar: "It is harder to watermark factual answers because the model has fewer alternative word choices available without altering accuracy."

In the codebase, I do see the models finding alternative word choices -- and I hate them. Already in a highly technical latent space, it reaches for other highly technical word choices (which may be more accurate, but ones I am unfamiliar with -- like terms in ERP systems as it felt that was close but bring just technical jargon that i have to google what it is saying because i don't understand -- and i have to google the phase sentence as the words themselves are ok, just not how they are put together).

sroussey··on Claude Fable 5.1 and Claude Mythos 5.1
My read was that watermarking is not explicit requirement, but that it could be in something like metadata that goes with it (if generated a word doc, for example). But writing comments in your codebase, the EU will want a digital trail there.
← PreviousPage 2 of 34Next →