vinzenzu

Moin ;)

I would like to use my Google AI Pro subscription included with Drive but last time I checked, their terms were not just ambiguous and confusing wrt training and ZDR, they were contradictory. Until they have that fixed and properly communicated, I'll stick to other vendors.

My biggest takeaway is Laya and that Jev isn't novel. I'll look into finetuning Laya.

oh my pi https://omp.sh/

There is no Opus 5.1. Guess you're talking about Fable 5.1

Qwen 3.8 Max was open-weight released [1], as was Qwen 3.8 Flash Next [2].

I still agree that they aren't as aggressively releasing the open-weights models as before, but there hasn't been a major release they haven't published the weights for yet afaik.

[1] https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B [2] https://huggingface.co/Qwen/Qwen3.8-Flash-Next

Btw, cool name ;)

We only offer SMAs, separately managed account, and only get a limited power of attorney to trade on your account. That means we can't withdraw any money, ever, and the account stays in your name.

So that's not possibel :)

We operate from Germany and everything must follow regulations and be compliant.

As you might have guessed from the sentence you quoted, we aren't managing client capital just yet. So it's not set in stone yet how we'll do it with the license. Our preferred path right now is with a partner we're in talks with which supplies the license (you can imagine it like a liability umbrella, Haftungsdach in German) and a respected German brokerage. Since it's not yet a done deal, I don't want to give more info. However, it's very important to me (Vinzenz Richard Ulrich) and will always be the case at autotradelab that we follow regulations and are compliant.

I've actually been considering talking with people (eg on forums) about the topic "asset managers suck" and finding out people's gripes. Since you brought up Bernie Madoff, do you have thoughts on that topic? We could chat on eg LinkedIn or email.

Buy & sell, and we focus on forex and crypto for now. In the future we'll add (index) futures & stocks. It's not LLMs trading btw. LLMs are not reliable for generating trading signals.

We use statistical and ML trading strategies, mostly statistical right now to be frank. Yet we constantly expand our zoo of strategies, as I like to call it.

There is no holy grail in trading, it's about having a system for market edge discovery (alpha discovery).

My co-founder is actually my father.

He lost all his clients when the brokerage he was an introducing broker for went bankrupt and then decided to focus on other things, like his (and his wife's) massage shop.

UNTIL he decided to return to finance for autotradelab, to bring his knowledge to a new generation.

Asset management using automated and AI-driven trading for portfolio diversification and higher returns.

Trading Results as a service.

Our philosophy is Capital protection first, so we focus on protecting capital and managing risk while trading, and avoiding large losses enables compounding and high returns in the long run.

Infused with my co-founder's three decades of experience in trading & finance.

We don't want all your money. We want to be part of your diversified portfolio, which should also consist of other things like index funds, bonds, etc.

Initial allocations will open within the near-term future. For the founding investors (as we call the first clients), we also offer a portfolio review to align your portfolio with your long-term goals, a direct line to founders, and other things.

https://autotradelab.com/

And here's the link on pi.dev's plugin/package marketplace

https://pi.dev/packages/pi-auto-reviewer

Codex and Claude Code have an auto/auto-review mode to let a reviewer LLM check suspect commands your coding agent wants to run.

I was annoyed that OpenCode and pi.dev don't support that, so I made this small plugin for the pi (pi.dev) coding agent to implement the same feature.

First plugin that does this for pi.dev that I could find.

It's been useful for me so I hope it's useful for you.

API prices of Anthropic, OpenAI, and Google are massively inflated.

https://martinalderson.com/posts/no-it-doesnt-cost-anthropic...

There's no way that all AI inference providers are colluding and/or all running at a massive loss, meaning the cheap Chinese model prices must be the real cost it takes to run frontier-class models PLUS their margin.

Look at Deepseek 4 Pro. https://openrouter.ai/deepseek/deepseek-v4-pro/providers Deepseek and Baidu are subsidising prices but they probably train on inputs. I have no model training and ZDR in OpenRouter enabled, and the first provider that shows up there is Deepinfra, significantly more expensive than Deepseek. BUT much cheaper than Sonnet 4.6 and ChatGPT GPT-5.4.

Any frameworks that I missed/that you're using?

We use NautilusTrader, very happy with it.

Leaps and bounds better! I don't think I benchmarked it.

But the experience was that it was able to find small details in PDFs, in technical diagrams, and this was really not captured well at all with OCR.

In general, OCR I think should be used more as an add-on to retrieve data, not given to the generation model itself. Similar to retrieving based off a text description and then giving the generation model the image.

For a RAG project for a client with a lot of PDFs and Powerpoints with images, I used ColPali a year ago. I see the provider ColiVara is still online but it seems to have fizzled out.

Retrieving based on text and then giving the generation model the image instead is much smarter than retrieving based on image. Image-based retrieval is slow and expensive.

Same with giving the model an image vs a structured representation of it.

Totally agree. I've been using cheap Chinese open-source models via OpenCode Go, and they are faster, cheaper and in my experience arrive at the solution quicker because they are more pragmatic.

Yesterday Codex was making a big issue out of a new module that was upgraded in our cluster and because of which the same SSH key would be "regenerated" by Terraform. No big deal, it just truncates a newline at the end of the SSH key and it works all the same. But not being aware that this, as an example, is unimportant can cost a lot more time than using the big models saves.