Yeah, I tried Codex pro today and the $20 plan is way more generous than Claude's, especially lately.

I've had the cheapest personal tier for both since forever and I think I've run out of Codex quota _once_.

With Claude it's a constant battle of typing /usage after every iteration and trying to guess if it's enough for the next task or not =)

I've found MiniMax 2.7 pretty decent and even pay-as-you-go on OpenRouter, it's $0.30/mt in, and $1.20/mt out you can get some pretty heavy usage for between $5-$10. Their token subscription is heavily subsidized, but even if it goes up or away, its pretty decent. I'm pretty hopeful for these openweight models to become affordable at good enough performance.

It’s okay, but if you compare it to eg Sonnet it’s just way too far off the mark all the time that I cannot use it.

I'm trying out codex for first time as well cause something up with Claude for sure, 4.7 has been super frustrating. For other models, highly recommend trying MiniMax 2.7, using it with Hermes is actually pretty good, and their token subscription plans include a lot of usage for $10.

Perfect, thanks. Codex app sucks, but I've been exploring opencode for that. Will try MiniMax!

+100 this. As devs we need to internalise this issue to avoid repeating the same class of exploits over and over again.

Worked on the software side of increasing the rate of solar penetration in electricity networks between 2016-2020 via global solar radiation forecasting. The uptake of the software was slow the first year but then rapid once more electricity networks were struggling with knowing how much solar was in the network. Once it is easier to predict, the network becomes easier to manage, and more can be safely added, and make it economically profitable. Sucks this was a commercial operation, but excited to see all the hard work across various industries is solving problems to get more renewable energy into networks.

Building websites, I agree has little value, but using it as a way to explain basics of how the web works I think is pretty valuable. Web likely isn't going anywhere for a long time, having some basic knowledge of how it works I think very useful for a lot of people. I hate the idea of any more MS apps like Excel being regularly incorporated, but basic usage of something similar definitely can help know of how to use a useful tool/computer skill. Even in the early 90's we had computer labs for learning computer skills which I think there is value. But forcing tech everywhere into teaching is an issue IMO.

"side loading", I know this term is the one used but I think should be pushed back against with just using the standard "installing"/"install". It makes the control point clearer and (should be) unsettling when you can't "install" software on hardware you own.

It's a great point. As a geek I used to think those details don't matter, but it turns out language shapes society and how humans think way more than I understood.

We need to catch up on this because the people who know how to use language for propagandizing don't have the best intentions in mind.

But using the original term is not enough. We need to combat their word-twisting by upping them. We need a way to convey "their way of installing stuff by default is inferior and an attack on liberty".

Something like:

- direct install: installing as we always did

- caged install: installing through a locked store.

Maybe somebody better at marketing can find a good way to do this. In fact, we should have a whole site and community to organize together and shift the narrative on all nerdy things: formats, open web, DRM, patents, etc.

We have been weak on these points for so long because we care much more about solving tech problems than selling them. But openness is being eaten away under our noses. Has been for years.

Sideloading should be called installing, and installing from the store should be called jailloading.

Jailoading is quite catchy, although it does have a "Micro$oft" and "Microslop" feel. Like more an insult than a word made to be used daily.

I think sideloading is a fine term when it is a consumption device. No one buys a video game console expecting to be able to install anything they want. As a matter of fact, there is an argument that restricting what can be installed is a feature. By maintaining control of the hardware, they can eliminate entire classes of problems that someone might run into. That is to say, when you let your kid play on the switch, you don't want to have to troubleshoot how they got the thing borked from installing malware.

That said, I do think words matter and I always point out that the reason these systems are locked down is because of Digital Restrictions Management. I also refuse to buy anything from Sony because they changed their mind about letting me install linux on the PS3.

I just think side loading is good way to describe installing custom software on a non-general purpose computer, and that not every computer needs to be general purpose. It's significantly better than the previous terms of hacking, cheating, stealing, and voiding your warranty.

Agree. I recommended Stremio to a friend on an iPhone and it turns out it has to be "side loaded". My response is "so you can't install it?"

I tend to draw a distinction - side loading usually infers a supported but not mainstream way of installing applications - this xbox for example cannot side load without you paying a small fee to enable the developer mode, and the vast majority of software will be obtained via retail discs or the Xbox store. It's not a generic "install" mechanism native to the out of the box experience for the console - you have to do some extra work for this avenue to open.

When I think of "install" I think of general purpose OSes which can install software from almost any source no questions asked, or use the native out of the box support for software installations.

The similar distinction exists with android and iOS, and is probably why the term is popular in those communities too.

If nothing else, the term sideload makes very clear on platforms with native appstores or locked down distribution channels (consoles, phones...) that the install did not come from the native channels. Installs from game discs or the xbox store are inherently different from developer mode software and using the same term "install" for both disguises this fact.

Yeah I listened to a podcast with Corey Doctorow (inventor of the term "enshittification") and he made this point quite well, to the point where I have completely removed "side loading" from my vocabulary. It's installing software on the computer I own.

Rugged individualism for the poor and vulnerable, won't someone think of the company and shareholders! /s

I think it would assist in exploiting exploring multiple solution spaces in parallel, and can see with the right user in the loop + tools like compilers, static analysis, tests, etc wrapped harness, be able to iterate very quickly on multiple solutions. An example might be, "I need to optimize this SQL query" pointed to a locally running postgres. Multiple changes could be tested, combined, and explain plan to validate performance vs a test for correct results. Then only valid solutions could be presented to developer for review. I don't personally care about the models 'opinion' or recommendations, using them for architectural choices IMO is a flawed use as a coding tool.

It doesn't change the fact that the most important thing is verification/validation of their output either from tools, developer reviewing/making decisions. But even if don't want that approach, diffusion models are just a lot more efficient it seems. I'm interested to see if they are just a better match common developer tasks to assist with validation/verification systems, not just writing (likely wrong) code faster.

I'm quite surprised the A100 is not much better since the power levels for the Ampere cards I believe is a lot lower.

Does this mean even for a model that fits on a single server that trains for a few weeks will absolutely need a recovery process? Interested in peoples experiences around this.

GPU servers always have had crap reliability compared to a normal server (but sticking eight GPUs on a baseboard complicates things). As I understand it (not my domain), this (being a lack of widespread checkpointing and mpift support) is one of the motivating factors for why ML toolkits eschew MPI (besides accelerator-accelerator being an afterthought).