What most AI model users don't understand is (most here will know) that they are probabilistic (statistically based) and not deterministic (static/fixed).
So answers to the same prompts can vary (wildly sometimes)
This really hit home yesterday as I made a general conversation state logging prompt (because suddenly Claude - Sonnet - was acting far more dumber than expected after /compact), and in 3 different projects it led to Claude saving the logs in 3 different locations (memory, .claude/session_state/.md, docs/.md). Somehow still didn't help increase the smart much BTW, so I'll probably have to engineer it, and just toss /compact out the window.
Honestly, that sounds more like you triggered the hidden "guard rails" or realignment (aka corporate censorship), and you got downgraded.
Of course, you have no way to he absolutely sure, cause Anthropic lies about silent downgrades, and charges you as if they didnt. And the 'realignment' rules are kept hidden, so you can't even avoid them intentionally.
Yeah it's very possible I hit a rail, as one of the projects involves fully automating Claude Code to allow co-driving (similar to what mobile remote control offers, but other agents (eg. Pi) can also be in control, and vice versa). And yes it does seem as though the token burn is also higher, even though it eventually outputs something that works). I'm a patient person though, as long as the rails aren't too restrictive and the subscription value remains reasonable.
Id recommend getting and running Qwen-3.5-35B-abliterated locally.
Having a local LLM is immensely helpful especially if you do the thing the companies disapprove of. Abliteration guarantees it wont refuse or censor.
The token dealers are incentivised in burning as much tokens as possible, and scamming by crippling/downgrading models. Their goal is to get you to pay more for each pull of the slot machine. And this slot (slop) machine doesn't have state/federal regulations for fairness.
Oh yes that reminds me, I have a few small local models including Qwen3.5. The tps rate is pretty bad for interactive use but I should definitely try it for my more automated workflows. I already use DeepSeek (via OR) alongside though and I doubt any Chinese provider will be restrictive in that way, so let Anthropic do what they may :).
I agree there... I hate the posts that rattle on about non-real-world work. Real world now (for me anyway) is maintaining legacy code, fixing issues, and new development. Now, some of the problems others have with AI are super crappy prompts with no real info other than "do this". I look at myself now more as a software designer and an architect, and the AI as a paid consultant. I give detailed specs and designs. The models are getting better at real deep code and legacy, but there are still holes. BUT give the AI a fighting chance with real detail in the prompts.
Yeah, I do this as well. Some of the last large tools I built have been started with Gemini (I get it with my Google package), and even one with Copilot.
I stopped using any of the built-in skills and skills agents and tried to keep things in the model only, and that seems to work better for token usage for me. It took me some time to tune my skills .md, but it rocks on now. I use them all in both Claude-code and Codex. I will say, GPT is better on token usage than Claude is, but for some of my code bases, Codex is not as good, but seems to be getting better with 5.6-sol.
Try to mix both subscribtions and take the best from both providers. In my opinion, we should not just one over another but understand the power of both and mix them to reach the better results.
Small or modular reactors in the US are more than 10 years away, probably more like 15-20. These are facts and not made-up political or pipe-dreaming techno-snobes.
> Small or modular reactors in the US are more than 10 years away, probably more like 15-20
Could be 5 to 10 with $20+ bn/year in scale and research spend.
Trump is screwing over his China hawks. The anti-China and pro-nuclear lobbies have significant overlap; this could be how Trump keeps e.g. Peter Thiel from going thermonuclear on him.
I work in the sector and it's impossible to build a full-sized reactor in less than 10 years, and the usual over-run is 5 years. That's the time for tried and tested designs. The tech isn't there yet, and there are no working analogs in the US to use as an approved guide. The Department of Energy does not allow "off-the-cuff" designs for reactors. I think there is only two SMRs that have been built, one by the Russians and the other by China. I'm not sure they are fully functioning, or at least working as expected. I know there are going to be more small gas gens built in the near future and that SMRs in the US are way off.
Guessing SMRs are a ways off, any thoughts on the container-sized microreactors that would stand in for large diesel gens? My impression is that they’re still in the design phase, and the supply chain for the 20% U-235 HALEU fuel is in its infancy, but this is just based on some cursory research. I like the prospect of mass manufacturing and servicing those in a centralized location versus the challenges of building, staffing, and maintaining a series of one-off megaprojects, though.
> it's impossible to build a full-sized reactor in less than 10 years, and the usual over-run is 5 years
I'm curious why that is. If we know how to build it, it shouldn't take that long. It's not like we need to move a massive amount of earth or pour a humongous amount of concrete or anything like that, which would actually take time. Then why does it take 15 years to build a reactor with a design that is already tried and tested and approved?
Well, you do have to move a lot of earth and pour A LOT of concrete :) Many steps have to be x-rayed, and many other tests done before other steps can be started. Every weld is checked and, all internal and external concrete is cured, treated, and verified. If anything is wrong, it has to be fixed in place (if possible) or removed and redone. It's a slow process and should be for many steps.
One of the big issues that have occurred (in the US especially) is, that for 20+ years there were no new plants built. This caused a large void in the talent pool, inside and outside the industry. That fact, along with others has caused many problems with some projects of recent years in the US.
When you're the biggest fossil fuel producer in the world, it's vital that you stay laser-focused on regulating nuclear power to death in every imaginable detail while you ignore the vast problems with unchecked carbon emissions and gaslight anyone who points them out.
i don't and i honestly don't know much about it, but
> there are no working analogs in the US to use as an approved guide
small reactors have been installed on ships and submarines for over 70(!) years now. Reading up on the very first one, USS Nautilus, "the conceptual design of the first nuclear submarine began in March 1950" it took a couple of years? So why is it so unthinkably hard 70 years later, honest question? "Military doesn't care about cost" is not good enough, there are currently about >100 active ones with who knows how many hundreds in the past, so they must have cracked the cost formula at some point, besides by now we have hugely better tech than the 50's, so what gives?
Yeah, I wondered about seacraft reactors myself. I think there are many safety allowances for DOD vs. DOE. The DOD reactors are not publicly accessible (you hope anyway), and the data centers will be in and near the public. There are also major security measures that have to be taken for reactor sites. You have armed personnel before you even get to the reactors, and then the entrances are sometimes close to one mile away from the reactor. Once there, the number of guards and bang-bags goes up. The modern sites kind of look like they have small henges around them (back to the neolithic!) :)
I created a local-only web app to wrap up some of my favorite web haunts, with HN being one of them. It allows me to look at the headlines, and save any of them in a locale SQLite db that the app maintains.
