Incipient

Better to have people suspect your model isn't as good as you're claiming, than to prove it outright?

I just wanted to say thank you for posting this. It was a wonderful break from the norm, and I've enjoyed reading everyone's comments. I haven't enjoyed anything here or in general in a while.

Thank you!

With luna's pricing, is the power consumed not likely to more than the cost of API calls?

At 50tps on a 3090 for qwen 28b or similar, that's about 200k tokens/hour. Which is what...10c for Luna (balanced input vs output). 1h of power at 30c/kwh is 30c? So you get a worse model, for more cost?

I think the opus5.5 vs fable5.1 situation has been my tipping point where I completely ignore these benchmarks now. Fable is just better than Opus5.5 in all cases I've ever used both and compared. I don't care what the benchmarks say, they're just not reflective of my needs, so they're not relevant to me - other people may have different experience.

I suspect a lot of organisations are already on automated AI detect-patch-redeploy loops for a lot of these types of issues.

Is there a definition of standard of living? A king would be entertaining, socialising, and strategising most of the time (along with a lot more pursuits of hunting, etc for leisure). While I understand they had less access to healthcare and variety of food, travel, etc...id still argue they had a better standard of living than someone that works 8-10h/day and then comes home to cook and clean?

I mean I ask Claude "please be concise in your responses" and it cuts down it's long response rate by a lot.

Using free copilot, which honestly is not the smartest model+harness out there, correctly identified to grab the ball and put it next to the elephant. I don't know what llm intermediary they're using here, but they seriously cheaped out...

>switch the defense model to using the same technology

Even the most sama-favourable person would struggle to look at that without even somewhat seriously thinking he's just trying to push even more AI usage/demand.

I'm not saying he's wrong necessarily. I'm just saying I don't read into someone selling AI saying we need more AI.

Side point from going to pypi:

> Security notice: Upgrade certifi, pip, and uv in response to CA certificate incident

Blimey PIP is affected? That's bad. I feel like that should have been bigger news?

I definitely feel something is "wrong" with these benchmarks.

I don't have much experience with anything outside of Claude for frontier models (and only haiku/3.x flash/5.6+6 luna at the entry level) so take this with that in mind.

I find Fable is still SIGNIFICANTLY 'smarter' then opus 5.5 when figuring out HOW to do something that's non-mechanical (e.g. How should we handle this aspect of business logic?"). Opus comes up with perfectly usable options, however Fable more often comes up with options that are more aligned with what I expect, or accept.

You don't/can't. You have to trust the server is sending you trustworthy js.

The only more tamper-resistant approach would be a browser extension, or a native application. Both have much higher friction.

We do. I think that's why it was 'rebranded' as gaming, not gambling/betting.

Anyone would be insane to pay for Adobe pdf over pdf xchange, but plenty of companies do. I'm not sure how/why the world pays for Adobe with so many good options but they do.

My personal view is on minimisation, not elimination. Some things practically must* be plastic. Where they don't (like microwave/storage containers, pre-wrapped vs loose item, etc I opt for the plastic-free version.

*economically/practically, not absolutely

Honestly I half expect these "food industry is bad!" videos/people are sponsored by the food industry itself. They're unhinged enough that I feel a lot of people dismiss them as just scare tactics. Unfortunately the stuff that goes into some products genuinely ARE in desperate need of being reviewed/banned.

These types of pieces I feel however do more harm than good to that cause.

I'm more referring to the colloquial 'separation of church and state', or formally "Congress shall make no law respecting an establishment of religion, or prohibiting the free exercise thereof".

Honestly if Somali pirates legitimately only attacked foreign fishing vessels breaching their sovereign borders - good for them! (I mean, it's quite brutal, but don't fish in someone else's water).

However, Somali pirates attack pretty much anyone, even those not disrespecting their sovereignty, which is less cool.

A few things I'd challenge here:

- the US believes religion doesn't belong in government, a lot of the rest of the world does.

-I don't believe in executing people for what they say, however I do also believe in a countries right to self govern, but I also believe other countries should condemn certain behaviour - it's a fine line with, and a fuzzy one.

- I also believe in 'when in rome'. Eg I was a Catholic in Morocco, and you can bet your boots I didn't go anywhere near their (fairly impressive!) mosques (as I didn't have a good understanding of their requirements, and I wasn't around long enough to have the time to learn).

While "stop making stupid people famous" doesn't quite apply here, "stop making stupid posts famous" probably does.

After he clarified and recanted a HUGE number of implied points in the first comment, my take is it comes down to "currently used encryption may be vulnerable" - essentially an evergreen statement, either due to weakness in the implementation (eg weak iv) or weakness in the maths (eg new mathematical attacks). Neither of these are new or honestly even worth pointing out.

Other than crypto with immutable hashes, the rest of the internet would simply update their systems to mitigate any broken encryption (depending on the risk of the 'break').

TLDR: scary implications make a post popular, regardless of substance.

>People like Euler

You met the chap?

Jokes aside, and regardless of how anyone perceives their field or industry, I feel pretty much every thought-based industry is up for an existential threat to some degree. Given a bit of time, manual labour fields are looking a bit shaky too.

The real question is not "are we impacted?" but rather "to what extent, and how do we integrate it?"

>incredibly simple to build the API and a background worker process with Effect. I get logging, telemetry, error handling, concurrency, API doc generation, and more for free!

But that can't be a library? Must be a framework? (eg more batteries included)

There is growing without regard to the lower 95% (china, India, USA) which is great on paper, but arguably not great for the population. I'd say the EU, Australia, and a few other countries have done it significantly better - it's just a significantly harder approach to balance.

Edit: and probably not as effective, which risks being overtaken by others, I suppose.

Has anyone tried these security models for finding bugs from the outside vs say Fable reviewing code for bugs on the inside?

Personally I'm a big fan of the EREV (which only drives an alternator, rather than a PHEV which drives the wheels directly) due to the simpler mechanical design - however I'm not sure if it's a personal preference. I feel even a fairly small 30kw (40hp?) would make a huge difference to the utility of the vehicle, while still being small and economical.

Not disagreeing with you, but also two wrongs don't make a right. Just because someone else, or someone does 'thing A' that's disagreeable, doesn't mean doing 'thing B' becomes fine.

Yeah we have too much urban sprawl, and too few people, to justify widespread public transport to eliminate the need for a household car (in Australia).

I don't think tbis quite holds - money being spent on domestically produced vehicles should be going to working families, so the money stays and is spent again in the country. Now if most of that money just goes to the mega rich, then it's fairly pointless.

But it's definitely more complex than just money saved = money spent elsewhere

I'm using fable and opus about 25%75/% respectively. I have a 3 step process - discuss a topic and create a review, convert the review to a task file and update the design docs, then finally implement the task file in code. Normally around 300-1000 loc implemented for a change, more for a new UI page or whatever.

And honestly it's incredibly good. - 50% of the time it'll generate code that I accept wholesale.

- 25%: it maybe makes a choice I don't like, or the implementation isn't the approach I liked, so I tweak it.

-10%: it needlessly creates semi-duplicate function logic, so it's a lot more code and execution paths become hard to conceptualise

-10%: it mirrors business logic it shouldn't, and subtly changes it. Which needs a huge amount of intervention to fix.

- 5% it goes completely off the rails. Eg decided that business/application logic should be in the database.

Prompt and check this honestly works very well (still waiting on tests with more than a handful of paying customers however!) but a FULLY automated agent pipeline? I'm not convinced. That last 15% of cases will make a degenerate codebase real fast.

Also without a CC subscription, paying per token, it would be horrifically expensive for my approach.

I'm not sure if agents are better or worse than humans at refactoring.

When they get it right, it works well - full context covered, full business logic recreated completely.

When they fail, they fail fairly hard. They miss some functions in a grep, or return too many and fail to address one part, and they end up with holes.

I scrutinise the work heavily, and find I rely on gut feel a lot for what's right and wrong.