JojoFatsani

Well, not everyone.

Opus is an Anthropic/Claude model, not an OAI/Codex model. I’m not sure if you can actually use an Anth model on Codex these days, but my experience to date indicates that performance tends to be best when you match the model to the harness it’s intended for.

You might also consider using the mid tier model (opus, Terra) to generate a detailed plan and then use a lower tier model (sonnet, Luna) to actually implement each task in the plan.

Probably everyone’s cd’s from back in the day are bit rotted or scratched.

Better public transit too.

Damn I had this idea but you beat me to it

I barely check them during the week

I just love that it knows how to form its thoughts into meaningful statements without rambling again.

Thou doth protest too much

Don’t blame the agent for not giving it the right requirements and skills to accomplish a task.

HVAC

Have done both and I'd say ECS wins almost always, as long as you can be in AWS. If you are true unicorn/hyperscale, Kube will start to pay off. But otherwise, the upgrade treadmill of kube and the blank stare of SWE's who don't want to learn helm charts, argo, ingress controllers etc are not worth dealing with when you can just ship ECS and have it work for years with minimal overhead.

startups/small-medium businesses implementing Kube are doing it for cargo cult reasons or stealth career development/employment security initiatives by the relevant devops or sre teams.

ECS fargate is a must as well because no "server" means a ton less compliance burden.

Why's that

This is one of the first things I asked GPT2 to do, and it came back with an awesome syntax and set of phenome rules which made an awful lot of sense to me. It suggested some particles or suffixes or something to indicated the kind of things that people use emojis for online (eg indicate sarcasm, empathy, etc).

Creating the language is of course the easy part, you have to get people to speak it.

100 years ago was 1926

18 years after 1926 was 1944

Solve for X

Spanish is very tokenized, feels like a good fit for interacting with LLM's.

The kid being a male isn't a reason he can't get a job at mcdonalds or driving uber.

Now do XP!

If there was no 9/11, there would be no carte blanche from congress to Bush to send the military off on whatever crusade.

It's smaller though, until you unfold it

The movie certainly is (directed by same guy as Robocop) I can’t comment on the source material.

I loved the idea of the charger Daytona when it first was announced, having been a fan of old school muscle but thoroughly in the cult of the EV, plus it’s a hatchback. But the execution on this car by Stellantis was just terrible. It’s boat sized, the artifical engine noise is corny, buggy electronics and it’s twice as expensive as it should be. I wonder how much they just pulled the rug from the engineers when it was clear there would not be more EV incentives.

If we are very lucky, the current Charger platform will have the same pushing 20 year lifespan as the last and the bugs and stuff will eventually get worked out, but it’s more likely everyone just buys a Model 3 or BMW NK than going anywhere near this thing.

The money is still good compared to breaking rocks.

Couldn’t you just use Karpenter for this?

Guy is crashing out. See what happens at the midterms if you can’t buy a Toyota or a Samsung.

It would be a fun exercise to use Claude to generate a spreadsheet of US exports designed for maximum reciprocation against the trump-voting/red senate states.

Because China has 3x as many people and they value education and social services

Why did you omit “would be” from the headline?

I mean, we used to just type doom.exe in our terminal

In the words of Cris Carter: Always have a fall guy.

This is more or less moot with LLM harnesses available to precisely manage the translation and mapping, or even just the ability to run the exact same cobol in a container for example.