phillc73

Deepseek 4.1 Flash 1017 has been pretty good for coding, I agree. It doesn't seem to do documentation as well, things like requirements, specifications, implementation plans. Here I still use GLM5.3 for a better outcome. I also use both models as adversarial batteries to check the work of each other.

Having said that, 1017 has been great in saving money. It's available generally much cheaper than GLM5.3, and I think it delivers >90% of the capability.

> Ends in 'a' = female: a relatively safe assumption, as far as an assumption here can be safe, with no other context. A strong majority in many places will match that.

Which places? I guess I ask because your next sentence:

> Many of the exceptions will be obviously foreign names with traditional spelling

Which makes me wonder if you meant "US places", because I can think of a few male names, which maybe categorised as "foreign" if you're in the US, that end in "a".

e.g. Csaba (pronounced "Chubba") is a reasonably common Hungarian name.

And some not so "foreign" ones, albeit mostly with Biblical roots, like Joshua and Ezra.

Which is a shame, because I'm sure there are some really good projects using GenAI effectively to produce high quality products. It's just that they're being drowned out by the slop cannons.

Some of them may be cider only varietals, which usually aren't super tasty on their own. This doesn't make them any less valuable, they just have a different use case.

For example, UK cider varietals include:

Dabinett (Bittersweet, high in tannins, well-balanced flavour, single-varietal and blended ciders)

Kingston Black (Bittersharp, high acidity and tannins, rich flavour, vintage-quality single-varietal ciders)

Australia, New Zealand, Ireland etc

Naming things is hard. Raven, modern scientific computing for OCaml has been around for a while: https://raven-ml.dev/

I use Maki[1] harness and GML-5.3, GLM5.3 Flash and Deepseek-4.1 Flash 1017, with a little bit of MiMo-2.6.

The key I've found is a strong skill. At the beginning, sure there were problems. However, I implemented an AGENTS rule that said if a Janet script failed more than twice on the same "gotcha", the correct syntax was added to the skill. The skill is regularly reviewed, consolidated and pruned. The Janet skill is a floor and loaded at the start of every session.

Edit: Also worth mentioning is that I developed a Janet module with runtime type guards and other helpers to encode the "gotchas". https://github.com/jennystats/janet-kit

[1] https://maki.sh

The beautiful thing about Janet is how lightweight and portable it is, no JVM.

I use Janet for all my agentic scripting. Python is forbidden in my harness. Janet scripts for everything. Love it.

I remember wild mushrooms, yabbies, eels as well. Prawning with my grandfather in a saltwater lagoon using hand nets and a hurricane lantern.

Did you eat them as a kid too? Growing up in rural Australia, I regularly ate rabbit in the late 70s and early 80s. It's been a while since I've had one though.

Growing up in rural Australia I remember eating rabbit as a kid in the late 70s and early 80s. One of my favourite meals was a sweet rabbit curry with rice that Mum used to make. Haven't eaten it in probably close to 40 years and my salivary gland is activated right now as I type this.

I flew Thai from Europe to Australia last month. The Economy legroom and space was pretty good. I did not feel cramped.

I’ll add Melious.[1]

I feel like I’m getting good value out of my subscription and they have the cheapest GLM-5.3 API rates I could find in Europe. I moved to them recently from Mistral for the better prices.

They do have a slightly different daily inference capacity allowance, but it’s actually not too bad once used to it. I was forced to up my game and figure out what GLM-5.3 Flash and Deepseek-4.1 Flash could actually do for me, rather than just throwing everything at GLM-5.3. Quite happily chugging along with a nice balance across them now.

I am hoping Melious make the new DS 4.1 Flash 1017 available soonest.

[1] https://melious.ai

I don't know if the MOPD patch still loops. I do know that whichever model Xiaomi is running on their own API as of two days ago does still loop (it's not clear if they're running the MOPD or not, but it would seem odd that they wouldn't).

> This is the first near-frontier model we've caught in infinite reasoning loops in a long time

Depends on where you draw the line for "near frontier model", but the current leading open-weights model, MiMo-v2.6, has exactly this problem. They released a fix on Hugging Face[1], but as of two days ago the model available via Xiaomi's own API still sticks in reasoning loops.

[1] https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Pro-MOPD

If you don't know Slopcore and Claudia, these links will help:

https://www.youtube.com/watch?v=C3fxudvU-UU

https://www.youtube.com/watch?v=kkwXVExCjg4

So disappointing really. Won’t someone think of the traditional values.

That’s right. Wise works with partner banks. However, their accounts are still covered by up to $250k of FDIC insurance.

And I’m in Austria. Plenty of businesses still ONLY accept cash. And not just small rural concerns. Had lunch in the city yesterday and the restaurant only accepted cash payment.

And the problem is people stopped paying cash, thus the facility was largely withdrawn. I don’t know what the answer is, but my gut says if cash remains legal tender, businesses should be obliged to accept it as a form of payment.

Otherwise, vote with your wallet wherever possible and prefer those businesses that do accept cash.

I applied to open an N26 account, and after completing all the necessary documentation they then told me I had to install their app to activate the account. Forget it.

Instead I opened an account with Wise, and have never once been forced to use their app. One occasion where some ID verification process pushed me towards the app online, I spoke with support and everything was sorted without it. Wise.com, just need a web browser and a phone number, zero phone app dependency.

Go even further and carry cash! No technology dependency and "just works".

As far as I understand it though, the Team has an overhead cost. e.g. 2 x individual Pro subs = €18.44 x 2. The Team sub = €29.99 per user per month, min 2 users. Their subscription page just says the Team plan has "even higher limits."[1] What does that mean? How much higher for the extra cost each month?

Essentially you are right, I could stack subscriptions via a Team plan, but for a premium.

[1] https://mistral.ai/pricing/

> ...but in the longterm the house/bookmaker always wins.

Not always. Yes, the house always wins in casino games like blackjack or slot machines. In sports betting, if you're good enough to overcome the market overround (the house margin), then you can consistently win and professionals have been doing this for decades. Read about Bill Benter in Hong Kong.[1]

> Kalshi and Polymarket now offer others the opportunity to become the house and take the other side of the bet against these people

This is not a particularly new thing. Betfair[2] has been offering this in the UK for at least 20 years and later in Australia. It's mostly sports betting, and of course Kalshi and Polymarket are not merely sports betting, but the ability to offer lay bets to the public is not new.

[1] https://en.wikipedia.org/wiki/Bill_Benter

[2] https://www.betfair.com/exchange/plus/

> is this British English for “they didn't follow anything at all”?

Yes, but the subtext is even stronger.

If you want to use a database instead of Polars, for similar use cases, duckdb is a much better bet than postgressql or sqlite.

Yes, but I still don't know where the truth is.... The Mistral dashboard shows a drawn down on €255 worth of "credit" when you have a Pro subscription, which costs €18.44. There's no way I'm going to use the pay-as-you-go API if it means I'd be burning €500 in the next six days, as I just ostensibly have in the last 6.

I appreciate I can set a monthly spending limit for the pay-as-you-go API, but I'm just not willing to find out how far €100 will go, when I know it will go further elsewhere.

I just wish they had that €100 subscription tier, for the equivalent of €1k pay-as-you-go use.

And Mistral also do exactly that. They host GLM-5.3.

I'm at a loss as to what to do now. I've been wanting to support Mistral for so long. I struggled on with Mistral Medium 3.5 for longer than I should have (although I also think it taught me some valuable process lessons).

Recently I switched to the Mistral hosted GLM-5.3, this worked very well and powered through a tonne of work. Unfortunately, I also completely maxed out two subscriptions within the space of six days this month. One can't stack subscriptions with Mistral, so I'd have to register a third account for another subscription, which will be annoying with changing API keys all the time. Sure I can switch to pay-as-you-go API, but that adds up really fast. The Mistral dashboard shows that a Vibe CLI monthly subscription for €18.44 actually provides €255 worth of API use (apparently, and I tried to check this with Support but it seems like they were intentionally vague).

After maxing out my Mistral subs this morning, I dropped $10 on Xiaomi to try MiMo-2.6. So far so good, seem to have done a lot of work for the $2.85 I've spent, and Xiaomi prices are still much better than the Mistral introductory offer for Le Chonk.

Not sure where to jump.

Edit: Not being able to stack subs is my biggest gripe with Mistral. I'd probably pay them $100 per month (5 subs worth), but I'm not going to switch to the pay-as-you-go API and burn much more money for the same amount of tokens. Instead, I've taken that extra money elsewhere. If they just allowed one to keep topping up subscriptions on the same account it'd be grand. Or even a bigger single subscription. Make a $100 tier with five times the capacity.