onlyrealcuzzo

5 ms·

31.
by onlyrealcuzzo 2mo ago
It feels like managing an egregious liar who speaks exclusively in corporate pyschobabble, is completely unpredictable, makes you want to blow your head off in most interactions, but yet is regularly good enough and cheap enough that you ca
32.
by onlyrealcuzzo 2mo ago
Step 1: use a different LLM that isn't 10x slower and 2-5x more expensive for the same level of quality.
33.
by onlyrealcuzzo 2mo ago
> I am curious why LLM writing has such an uncanny valley feel to it. Because they are HEAVILY trained to give addictive responses. They don't want to just answer your question. They want to sycophantically make you feel like a g
34.
by onlyrealcuzzo 2mo ago
Similar to the way they asked Sol to solve Erdos problems, that's what I want my model to do for programming. I don't want to try to take my best educated guess at what the best design is BEFORE implementation - especially if you&
35.
by onlyrealcuzzo 2mo ago
If the benchmarks don't lie, this is getting very close to Opus 4.6 capability - which was the turning point for me for when AI was "good enough" that it became very hard to justify not using it. I'm sure there's so
36.
by onlyrealcuzzo 2mo ago
Claude is terrible when it comes to speed. Flash is great, but Codex models are also fast, as is DeepSeek v4 Flash. Anyone who's Anthropic-pilled should really get out and explore and see how unbelievably terrible they are when it come
37.
by onlyrealcuzzo 2mo ago
I suspect Anthropic wanted the US gov to ban Mythos for marketing. If it turns out to be bad for them, the US gov will likely suddenly unban models.
38.
by onlyrealcuzzo 2mo ago
> 13-26x cheaper with comparable intelligence, and available across many different inference providers. Well, compared to 2 months ago, it's no longer 100x more expensive for similar levels of quality... If they continue monthly-ish
39.
by onlyrealcuzzo 2mo ago
Sonnet 5 is arguably the most cost ineffective model to ever be released, so that's not really impressive. It can regularly cost more than Fable, take longer, and deliver far far lower quality. I'm much more interested how this co
40.
by onlyrealcuzzo 2mo ago
2B is pretty small... No 100B+ param (certainly no 2T+ param) models have been trained natively to quantize down to 1.58b.
41.
by onlyrealcuzzo 2mo ago
~99% of my success is that I happened to make friends with the right coworker at my first job, who happened to be a genius, and happened to refer me to better places. Otherwise, I'd still probably be making 1/4th of what I'm
42.
by onlyrealcuzzo 2mo ago
Standard models are designed to quantize down to 4-bits relatively well. Anything below that, and especially 1.58b - is typically complete garbage, and you're much better off running a model 100x smaller at regular precision (compared
43.
by onlyrealcuzzo 2mo ago
Isn't this quite a bit behind Sol and Fable and even ChatGPT 5.5 xhigh and Opus 5 max? In terms of what you get for what you pay for, it's incredible - probably by far the best. But unless I'm reading things wrong, it does no
44.
by onlyrealcuzzo 2mo ago
I suspect it's because there's a flood of people writing / looking at MD docs that never did before and want better tooling than is readily known to exist. I think we're going to see a lot of dev-related tooling that tra
45.
by onlyrealcuzzo 2mo ago
What makes this insanely hard to predict is that the compute needed for the same quality output has roughly gone down 90% every 18 months for ~5 years. 1) We don't know how long that trend will continue, but you do know where to look f
46.
by onlyrealcuzzo 2mo ago
I would hope that Qwen 3.8 is better. It's been 4 months, and we've seen almost no progress in this space. As people have called out, Glimmer appears to be a trade-off rather than a clear winner. And from what I've been readi
47.
by onlyrealcuzzo 2mo ago
Isn't this a plateaued market? Amazon, Google, and Apple are already entrenched in the space. It's not like any of them have toxic brands like Oracle where everyone is dying to free themselves. Most people are happy with their p
48.
by onlyrealcuzzo 2mo ago
Does this work for different harnesses like Codex and Antigravity?
49.
by onlyrealcuzzo 2mo ago
> Is Amazon too broke to buy the necessary land to build out wind/solar/battery? It's more capital intensive, and interest rates are 8%, because our elected officials are completely incompetent.
50.
by onlyrealcuzzo 2mo ago
As someone working on a language in this space, I think your README could be much better. Most of the important stuff is in this section: ```3. What's Actually New About It?``` But you don't really go into any details about what&#
51.
by onlyrealcuzzo 2mo ago
> Deepseeks secret sauce is the incredibly cheap caching (magnitude cheaper than other providers). Can anyone working at one of the main US labs (Google, OpenAI, Anthropic) comment on WTF they haven't even tried MLA - despite the ob
52.
by onlyrealcuzzo 2mo ago
It's not going to for a long time because the cost/benefit isn't there... Even if Waymo's healthy growth rate continues, it won't make up 5% of the entire US rideshare market until 2030. That's still several ye
53.
by onlyrealcuzzo 2mo ago
No True Scotsman only drives in the city. All True Scotsmen drive across the entire U.S. of A. Waymo has completed 20 MILLION autonomous rides, 6.5m last quarter alone... At what point does it get to say it can self-drive? When there&#x27
54.
by onlyrealcuzzo 2mo ago
> Well, we don't know that, because neither have been achieved. Sure, we will completely ignore Waymo and move goal posts...
55.
by onlyrealcuzzo 2mo ago
I also don't think the analogy will age well. It might take more than 10 years, but robots will almost certainly master cooking single pieces of meat. It's substantially harder to reach self driving than it is to cook a "good
56.
by onlyrealcuzzo 2mo ago
Beef is particularly egregious. Poultry and even pork are about an order of magnitude more efficient. It's interesting how long the market was wildly subsidized to hide the true cost of beef.
57.
by onlyrealcuzzo 2mo ago
Way back when I started my career - I almost COULDN'T take a good job in Boston, because I couldn't figure out how I could literally live there long enough to get my first few paychecks to pay for my rent. I decided that I'
58.
by onlyrealcuzzo 2mo ago
Guys, it's fine. Relax. It's only a crime if someone else does it or is not sufficiently wealthy, otherwise it's a mistake.
59.
by onlyrealcuzzo 2mo ago
There is no other way to describe the Gemini 3.5 pro delay than as a complete and unmitigated disaster. It's a huge company. It's unlikely there is any one person to blame (and entirely possible he has none of it). But things need
60.
by onlyrealcuzzo 2mo ago
What could possibly go wrong with Dr Strangelove?

More ›