jannyfer
jannyfer on Hacker News
In the article they did send the navy.
I know the paper specifically calls out Opus 5.5 and Opus 5, but I can’t see different versions of Sol, or Astra.
I think the difficulty in comprehension was mostly an Anthropic problem… partly because Anthropic employees stopped reading Opus output around the time Fable came out and stopped noticing RLVR going off the rails with compression.
The paper was a good read, in any case. Maybe I gotta start sending this article back to LLMs when they start compressing text too much :)
> submitted Oct 5 2021
(In case people miss that before discussion)
Wow, 16 PRs?
No wonder GitHub keeps going down.
“When implemented correctly” is probably what people are complaining about.
Opus 5 started adding a bunch of comments to code, even when instructed not to, and for very simple changes where the comment itself was longer than the code change. Was that so that there are enough tokens outputted for watermarking? Many people suspected so.
I believe there are studies about how the environment and who you are with changes your perception of food as well (it was some Radiolab podcast episode I believe). So you will never again have such a great sandwich, even if the exact same sandwich was brought back to you now. But I think it might be good that way - the free flashback and remembrance.
I remember one of my own "best steak ever" moments. It just never hit the same, even though I've had steaks 4x the price since then. I've also had a "best baguette ever" moment and I'm pretty sure it was just a regular baguette but my brain released an incredible amount of dopamine.
On a similar topic - since you said "I always use" - is it true that nearly everyone at Anthropic is using Fable and that's why Opus 5's style of speech was not really noticed as problematic by Anthropic employees?
I agree that the LLM behaviors between Anthropic and OpenAI are different enough now that it can lead to different experiences.
I absolutely hate talking to Opus but don’t mind Sol and Astra.
My guess is Martin used Opus.
A lot of the more sane people I work with have decided that it's futile to try and have stopped working with Opus (we can't use Fable at work due to data retention policies).
RIP. I had frequently used ALT + 0151.
They have a {{currentDateTime}} in the prompt which is interesting for something that gets prefix cached. Hopefully they are handling that properly :)
Interesting, they say "focused, brief, and concise" but not "clear". I find the Claude responses these days are too dense, which contributes to being overwhelming. It's not the number of words used that overwhelm me. I've been asking it to use ASD-STE100 and that helps somewhat.
Hm? I'm saying that the AI firms used to have the philosophy of "ok this kitchen knife is dangerous but we'll catch the murderers" on older AI models. But now, the AI firms think that any average person could send a flying knife to attack a political figure they don't like, from the comfort of their home. Now give this to a billion people, and suddenly you have chaos. So to continue the analogy, now they're mandating drone registration, GPS tracking, etc.
And then a Chinese company sells a drone with no registration or tracking and suddenly people want to turn to legislation to ban Chinese drones.
Hey this analogy is working really well
For a kitchen knife this was okay, but the AI firms think that they’ve built a drone that’s the size of a phone but can fly 100km and can hold a kitchen knife. It might be used to assassinate someone before others can react or even catch them.
Fun article, but this didn't actually answer the "why" question for me...
Or be like me and just mulch the clippings
(yes it looks uglier that way - and yes I think of mowing the lawn as a chore)
The article's EEG images don't seem to suggest this is due to two brain hemispheres.
Yup, and I had seen it in the Opus 4.5 soul document as well:
https://gist.github.com/Richard-Weiss/efe157692991535403bd7e...In the future, I expect different models from different firms behaving differently will become as obviously normal as different humans behaving differently. Those of us that have used agents a lot have already noticed this, but the general public still seems to consider AI to be plug-and-play.
The Opus 4.8 infatuation with being honest genuinely drove me nuts (honest take). The constant need for me to decipher and decide on something after its final message was tiring too.
GPT-5.6-Sol is refreshing that its replies start with "understood." instead of flattering me for my steering prompt. Now ... after a few weeks, will I get suck of "understood"? I dunno :)
And the fact that this article’s story is basically “I prompted Fable with a goal and went to sleep and the model got it done” is telling me that the latest models have gotten past the need for Superpowers… even the creators of superpowers is just using a simple /goal!
Yeah I enjoyed the first bit, then “the one exception” made me go “hey, Claude does this to me all the time” and then it ruined the article for me.
To be fair, I saw that, I saw the “skip” link, and thought “that’s ok, I’ll just scroll down” on my phone.
It took SO many scrolls to get to the point. So the frustration is justified.
That was fun, and the script on github looks hand-written which is refreshing after having been reading AI-written code for months.
I have 120k photos in iCloud that I'm sure have duplicates (I exported my library to Google Photos years ago and exported it back to iCloud). The iOS duplicate detection stopped flagging duplicates for me to merge a while back. I gotta do something like this script...
I see this post is from 2024. Maybe I would have enjoyed the hook and enjoyed reading along to figure out what this "harajuku moment" was back in 2024. But since being exposed to AI slop daily, and having to scan through so much verbose AI outputs during day-to-day "coding", I've now started skimming so much that I got annoyed that it meandered, then just couldn't bother reading the rest of the post after I've figured out what the harajuku moment is.
It's like my brain is responding to blog posts now in the same way that people scroll past tiktok videos in the first few seconds if there isn't enough of a dopamine hit.
I used to enjoy longform content... alas.
Oh then that's like the battery swap idea but without the swapping!
This is how I used to solve the roomba-can’t-use-stairs problem.
I’ve now moved to a single floor. Problem solved!
Agreed, I used to think this but now enjoy taking quick selfies, and my phone will dig them up and remind me of fond memories later on.
GP conflates selfies with posed photos.