jeeeb

I’ve gotta say my experience has been the opposite. I’ve been consistently surprised at how much mileage I can get out of the $20 plan.

My usage is pretty simple. I just use codex with an AGENTS.md to point it in the right direction. I have no MCPs and use Sol on medium thinking as the main model.

For longer sessions I tell it to launch a sub agent to make changes. Otherwise I just reset my context for each new task.

And Japan isn’t either. They are requiring that by the time you retire you’ll have paid at least 30 years into the pension system or have equivalent financial resources any gap.

I don’t think this is a very good use case. You could do it better with a strong LLM and structured outputs.

The problem is that you want the model to carefully reason about the goal and code.

Zero shot classification with an approach like this isn’t going to do that. It’ll answer on first pass vibes.

Why? What business is it of yours?

They were probably under an agreement not to reveal the buyer or amount.

Meta in turn probably wouldn’t have wanted news to get out and people forerunning them to claim other online properties just to sell them back.

That cuts both ways. US entities own huge amounts of overseas assets including debt.

A more appropriate saying would be cutting off your foot to spite your leg.

Thankfully no one has done that yet.

We have a subscription at work. They have a CLI agent that works pretty well. Lately I mostly use it with grok, which is actually quite a capable model.

Aider (and probably others) came well before CC.

No. Japan certainly doesn’t. If you want to renew your visa you go apply at the local immigration department and they mail you your renewal notice a month or two later. No need to leave the country at all.

It’s not really clear what you mean. 100kWh is more than the entire battery capacity of the EV.

Agree. Another consideration is that input token costs are effectively quadratic with the number of conversation turns. The minute you get into a long conversation you can see the costs shoot up.

You can get far more gas out of even a $20 plan of you’re careful to break things up into relatively small discrete steps, clear context regularly and give the model plenty of information to work with.

My workflow for bigger features is to write out a plan document and then proceed in smaller implementation steps, reviewing as I go. If I find something odd, I ask the agent why and often that leads to discovering a new dimension to the problem, which in turn is an opportunity to adjust the approach.

Grid scale batteries and household batteries are being widely deployed.

Australia is the third largest market in the world for grid scale batteries, and has the highest per-capita capacity in the world; https://www.pv-magazine.com/2025/10/21/australia-becomes-wor...

Not to mention more than 200k new household batteries installed in 2025 (out of roughly 10 million households).

I have T1 diabetes and wear a GCM so have a continuous view of my blood sugar levels. I also go to the gym and exercise otherwise.

My experience is somewhat different. Even walking is more effective than chess presses or arm exercises for immediately reducing blood sugar.

After an hour or so of more intense aerobic exercise I find it basically impossible to avoid a hypo (low blood sugar).

Simplistically if the effect of exercise on blood glucose is roughly proportional to Volume of muscle engaged x Length of time engaged, things like running should be very effective. The bulk of your muscle is in your lower body and you’re continuously engaging it.

I suspect any decently intense exercise can have a moderate lowering effect on blood sugar levels after the immediate drop seen when you’re exercising, but the effective isn’t huge.

This is simultaneously amazing and horrifying.

I feel like we’re at the stage where if AI decides it needs to delete your production DB to solve the user login problem, then it’ll find a way to do just that.

If anything, it’s a very conservative estimate. Short of a major turn of events it seems very unlikely Anthropic’s revenue growth is going to slow to zero.

I think the government should make the new 0% tax rate optional. Stores can either continue to collect 8% or drop the rate to 0%.

I think that might help motivate these companies to move things a long a lot quicker. They might even find it doesn’t take 1 year after all.

Second/third generation children of Japanese immigrants to Brazil were historically given special visa access.

I assume the OP is actually referring to these returned second generation Japanese.

My guess is that it was an abuse filter over-keyed on OpenClaw pushing Claude to certain responses patterns.

Normally given a web search tool these models will search if they don’t know.

I said the model has access to a web search tool. Of course it’s the harness that provides that access.

I didn’t spell that out because it’s irrelevant detail, immaterial to the point I was making.

Knowledge cutoff is completely insufficient as an explanation.

These models have access to a web search tool. Gemini and ChatGPT both happily search for give info on OpenClaw. Claude denies all knowledge.

What’s more it’s this part that’s very concerning.. Banned for wrong think..

> I gave it a direct link to openclaw.ai and the chat instantly ended and hit my 5hr usage limit.

Is actually running the unit tests what is important here though?

Could you achieve the same by just telling the agent: “This is a Python project. We use uv and pytest”

It feels very much like Gemini’s writing style - overly excited with lots of unnecessary contrasts.

> It is also a bit weird that they are not incorporating speculative decoding

Wouldn’t speculative decoding decrease overall throughput, but optimise (perceived) responsiveness?

These look like good results for a first model release. I’m hoping to see more, especially in the 30b parameter range.

Japan giving a security guarantee to Taiwan would be major news!

In reality no such thing happened and one YouTube video of a handful of protestors doesn’t make it so.

What she did say is that a Chinese attack on Taiwan _could_ clearly become an existential threat to Japan. Note that key word _could_

Which… of course it could!

Japan hosts multiple US military bases. If it developed into an armed conflict between the US and China then it’s exceedingly likely that Japan would be attacked. Think Chinese missiles aimed At Yokosuka, just south of Tokyo.

Not only that but Japan and China have multiple territorial disputes. It’s not hard to imagine China deciding to go all in and settle those as well.

If this was just about semiconductors then this would be a reasonable take but I doubt semi-conductors are anything more than a minor footnote in China’s strategic calculus vis-a-vis Taiwan.

Reunification with Taiwan has been a major policy goal of the CCP since the civil war and is one of Xi’s explicit policy goals. He just reaffirmed this commitment as part of his New Year’s speech.

Historically China has lacked force projection capability. However it has had a multi-decade modernisation and military build-up which has drastically changed this situation.

Further we’ve seen significant tightening of CCP control over society and in particular the military in Xi’s term.

A straight forward analysis of these events, in line with Xi’s public statements and past Chinese actions, is that the ground work is being laid for encirclement of Taiwan followed by China taking over, by force if necessary.

Looking through this guys GitHub he seems to have a lot of small “demo” apps, so I’m not surprised he gets a lot of value out of LLM tools.

Modern LLMs are amazing for writing small self contained tools/apps and adding isolated features to larger code bases, especially when the problem can be solved by composing existing open source libraries.

Where they fall flat is their lack of long term memory and inability to learn from mistakes and gain new insider knowledge/experience over time.

The other area they seem to fall flat is that they seem to rush to achieve their immediate goal and tick functional boxes without considering wider issues such as security, performance and maintainability. I suspect this is an artefact of the reinforcement learning process. It’s relatively easy to asses whether a functional outcome has been achieved, while assessing secondary outcomes (is this code secure, bug free, maintainable and performant) is much harder.

> That feels like cargo-culting the toolchain instead of asking the uncomfortable question: why did it take a greenfield project to give Python the package manager behavior people clearly wanted for the last decade?

This feels like a very unfair take to me. Uv didn’t happen in isolation, and wasn’t the first alternative to pip. It’s built on a lot of hard work by the community to put the standards in place, through the PEP process, that make it possible.

What uv did was to bring it all together.

That’s 8th of June 2023 not 2025.. almost 2 years before Claude Code was released.

I remember evaluating Aider and Cursor side by side before Claude Code existed.

That’s what most languages, including Java do.

The problem the OP is pointing out is that some programmers are incompetent and do string concatenation anyway. A mistake which if anything is even easier in Python thanks to string interpolation.

The idea in Java is to let the JIT optimise away the logging code.

This is more flexible as it still allows runtime configuration of the logging level.

The OP is simply pointing that some programmers are incompetent and call the trace function incorrectly.