AustinDev

Let's say your company makes holes. The year is 2026 so you use excavators because they are good at digging holes quickly. Would you hire a guy who refuses to use an excavator but will happily use a shovel?

AI explainer videos are actually pretty good. Good usage.

There was a game that was Jev trolley problem PVP. The meta appears to be that you need to make the most mainstream journalism victim case. The more holocaust adjacent the better

https://trolley.typememetics.institute

It was an interesting experiment. I got to top 50 without a racial or religious based prompt but couldn't beat the top tier grievance folks.

Yep, just benchmarked it against GLM5.3 Flash on my own local bench and it's slower in TTFT, tok/s and concurrency. It also scored lower.

Tried different reasoning levels and didn't really see any difference but GLM 5.3 flash only really works high, max or none so I couldn't really compare the lower efforts which appear to work on Spark.

I lived through it. I've listened to the podcast. 2026 is not 2021 or 2022. Since about 2022 or maybe 2023 I have never received a single ERCOT warning message text. Before then they were pretty common.

I'm saying it was a problem, it's been addressed. We had one of our hottest Augusts on record this year without a single issue.

If you think it's structurally broken there are energies futures markets you could use to make a lot of money on.

From what I can tell a bunch of batteries, solar and wind came online since then not sure if the govt was involved or not.

The power has been great since the freeze. Market forces fixed the problems. People were making $9,000/kwh at times. That's never going to happen again.

And for that same reason people knit their own scarves but they don't expect the scarf they see at the store to be hand knit.

>You can already performantly run frontier level models on systems with $1600 tier video cards.

Quanted out Qwen 3.8 models are not frontier level. Have you used them for real work? They suck. They are so unsure of themselves due to the quant-lobotomy. They just endlessly loop in thinking even when they have the right decision and then output it when they get to the thinking cap. They score close in benchmarks because most benchmarks don't measure time to complete a task.

I use Qwen quants in production for fast inference but I disable thinking and have them fine tuned for the task. My realtime transcription + translation stack right now is running 5 streams at 380 tok/s.

Running DeepSeek 4.1 Flash at FP4 is going to need at least a 10-15k machine. I'd consider that near frontier level.

What are you even talking about?

What is the purpose of a nation if not to benefit its own Citizens. If it doesn't do that in good times then why would anyone fight when called upon?

I don't follow? Why would a US citizen want to compete against 8B people for jobs instead of just 300M? Seems like the lower number is better for the citizens of the US.

encode and decode tok/s which is ($/s) when it comes to pricing drops heavily above 100k tokens.

There are plenty of workflows like translations where you'd easily be under the cap.

This makes sense when you realize most ICE car makers get a substantial portion of their revenue from maintenance and replacement parts.

Fair enough. Cheers

Still happens on modern models like ChatGPT Image and Qwen 2.1. It's extremely rare in Gemini since nano banana's release. I work with people who generate 1,000's of images a week and it'll happen. There is a slack channel for it where you tag the model.

You're not going to know if they used an LLM to write a better shader or if they copied code from stack overflow like we used to do a few years ago.

I'm so excited that LLMs are great at coding. Most of my problems I have with various games I love is that they're written by people who couldn't write performant code at all but are great at building a world and telling a story.

Just look at Escape from Tarkov. Great game but some of the worst software architecture choices of all time. It used to take 4 minutes to get out of match and often 10-15 minutes to get into one and it was all just because of shitty code / design.

It's been fixed now and I'm sure modern LLMs were the reason.

John Carmack is the exception not the rule.

All of these things would require humans to prompt the models. So... the humans doing the prompting should suffer the consequences, this isn't that complicated.

I think railway safety and nuclear safety are fundamentally different from AI Safety.

I would bet the vast majority of the world population would agree that they don't want to see trains derail or nuclear plants meltdown.

I don't think there is that sort of agreement when it comes to the question of AI Safety.

Is generating the founding fathers of the US as Africans good AI Safety? To some people maybe.

Just put the code in the bag.

Sentiment is mostly correct, but the writing is terrible.

Starlink is too low to be a concern for Kessler Syndrome. They all deorbit on their own in a few years. 700-1,000km is the orbit that should be observed for Kessler risk. <500km is of no real concern due to atmospheric drag.

The US does own the moon for all practical purposes, or we did until we gave up our capacity to reach it.

Is it the public commons if you can't reach it?

and anthropic won't? or any other inference provider? Running your own inference either locally or remotely are probably the only ways to make sure that doesn't happen.

gemini 3.8 flash?

Hmmm, I've never actually checked that. Just had codex run the numbers excluding the boilerplate in some of the documents output is on average ~13% shorter than input, I only checked the ones that were all of the same type (proposals) since it was easier to have codex run the numbers on that.

I compared at final edited output less boilerplate vs sum(prompts).

If I had to guess I'd say far more people are better at editing than they are at writing de novo. My approach probably limits the creativity of the output but I'm not writing a novel.

Here's how I use it to write. I dump an unedited stream of consciousness into the prompt. Sometimes it's just me having a conversation with myself thinking through what I actually want to say. Then, I have it ask me to clarify anything that's unclear. Then I have it write me a rough draft. I read that and then just rewrite most if not all of it while I edit the output. I use this workflow for work stuff or dealing with businesses in personal life. I'd never use this to generate a letter or email to a friend or colleague.

Pure slop, completely unadulterated from what I can tell.

It doesn't really matter it's all the same thing for the point of the discussion. All investigations into these 'hacks' are focusing too much on the model and not enough on what the humans did wrong. The model doesn't have real agency it can't be put in prison so what it 'thinks' is irrelevant. We need to be focusing on what the humans did in these situations and assigning guilt based on those findings.

I don't fix typos anymore unless they change the meaning of what I'm trying to communicate. Don't want my human writing to be confused with LLM output.