AI Development general thread.

quovadis said:
In time. Imho it's inevitable that highly capable models become a commodity for offline usage.
Click to expand...
Yeah, I've tried most of the open source models via runpod and they just don't even come close to frontier performance.
 
InTheWild said:
geez I went through a whole rabbit hole with claude on a trading bot. Had a python app called Norman the trader running on a small VPS and connected to Telegram and did trading through the VALR API. Mixed results so far, although it wasn't doing any shorts and there were no gearing added. Technically it made a small profit even while the market was in down trend for the past few weeks.
Click to expand...

Yeah been using Hyperliquid works pretty well. Want to fine tune my openclaw then set up a Hermes (apparently its should improve better over time) and the go head to head in a trading challange with them to see how it goes.
 
quovadis said:
In time. Imho it's inevitable that highly capable models become a commodity for offline usage.
Click to expand...

Been using olama and qwen on my gaming pc, then openclaw using anthropic api to review and decide if any trade recommendations are worth it. Currenlty win rate at 50% but wins usually greater than losses. Still early days and things break a lot but fun litle side project so far.

Been busy since March and the improvements in AI over the past few months has been impressive, Opus 4.6 was great, 4.7 felt very slow and 4.8 feels like a decent upgrade again.
 
semaphore said:
Yeah, trading bots with LLMs... @cguy can tell you how that's going to pan out for you :P
Click to expand...
Yeah, not well. The analogy I would draw is that it’s like trying to compete with a car engine, by taking a humanoid robot and using it to power a pedal car.
 
Anyone want to evaluate or already tried out Gemma 4 12B?

ollama.com

gemma4:12b

Gemma 4 models are designed to deliver frontier-level performance at each size. They are well-suited for reasoning, agentic workflows, coding, and multimodal understanding.
ollama.com ollama.com
 
quovadis said:
Anyone want to evaluate or already tried out Gemma 4 12B?

ollama.com

gemma4:12b

Gemma 4 models are designed to deliver frontier-level performance at each size. They are well-suited for reasoning, agentic workflows, coding, and multimodal understanding.
ollama.com ollama.com
Click to expand...

Sill on Qwen 3 5:9b might give this a go should run well on my current setup
 
How are you guys in enterprise handling Tannie Sannie in finance who has now vibe-coded some magical new dashboard and needs to deploy it company wide but has no idea what a repo or merge request is?

This is something I'm trying to plan for because I expect the influx is just around the corner and in my mind something Portainer-like might be the answer and demanding that it be a container in the first place otherwise no dice.

Or do you handle the whole thing end to end and just take their code and deploy it for them and now it becomes a you problem?
 
SauRoNZA said:
How are you guys in enterprise handling Tannie Sannie in finance who has now vibe-coded some magical new dashboard and needs to deploy it company wide but has no idea what a repo or merge request is?

This is something I'm trying to plan for because I expect the influx is just around the corner and in my mind something Portainer-like might be the answer and demanding that it be a container in the first place otherwise no dice.

Or do you handle the whole thing end to end and just take their code and deploy it for them and now it becomes a you problem?
Click to expand...
+1 to this. I’m still trying to fathom wtf we have Tannie Sannie vibe coding but here we are. What a dangerous time to be alive.
 
PaulKempTheSecond said:
+1 to this. I’m still trying to fathom wtf we have Tannie Sannie vibe coding but here we are. What a dangerous time to be alive.
Click to expand...

Yeah and the problem is even with all the developers in the world you can't have someone go code review this with any kind of proficiency because you ask the AI the same question three times it will spit out three different languages for each...so you will have to rely on the AI tools to review itself.

At least checking it for secrets and PII isn't too hard, but yeah the whole thing is a minefield.

My logic is as long as there is some default authentication in front of it and it's purely internal then some bases are already covered and otherwise it will be treated as ephemeral by default. If it needs to live longer than it can go through the usual full production cycle.
 
Anyone using Fable 5 yet? Have it available on our plan but heard it chows through tokens. Might just try it to see what it's like.
 
DrJohnZoidberg said:
Anyone using Fable 5 yet? Have it available on our plan but heard it chows through tokens. Might just try it to see what it's like.
Click to expand...
Available on my plan till June 22 before it starts using credits.
Used it briefly earlier in the week to check a bit of python. Gave me wrong answers so did the same in copilot which picked up the problem.
 
SauRoNZA said:
How are you guys in enterprise handling Tannie Sannie in finance who has now vibe-coded some magical new dashboard and needs to deploy it company wide but has no idea what a repo or merge request is?

This is something I'm trying to plan for because I expect the influx is just around the corner and in my mind something Portainer-like might be the answer and demanding that it be a container in the first place otherwise no dice.

Or do you handle the whole thing end to end and just take their code and deploy it for them and now it becomes a you problem?
Click to expand...
We already have this happening. You tell them to create a package, apply for firewall rules and push to ops for deployment. They must also be present for acceptance testing over release weekend to fix any bugs with their app.

It quickly turns into "ek soek help met my app asb". The bigger issue is actually they have no idea how to align with architectural patterns and integrate into existing systems...
 
Mike Hoxbig said:
We already have this happening. You tell them to create a package, apply for firewall rules and push to ops for deployment. They must also be present for acceptance testing over release weekend to fix any bugs with their app.

It quickly turns into "ek soek help met my app asb". The bigger issue is actually they have no idea how to align with architectural patterns and integrate into existing systems...
Click to expand...

Yeah that’s why I’m thinking to keep this kind of thing entirely separate that can spin a container up and down on demand.
 
DrJohnZoidberg said:
Anyone using Fable 5 yet? Have it available on our plan but heard it chows through tokens. Might just try it to see what it's like.
Click to expand...

Used it last night, seemed decent but hit my token limit in 2 hours. Worth it to test till the 22nd or whenever the plan doesn't cover it anymore.
 
SauRoNZA said:
Yeah that’s why I’m thinking to keep this kind of thing entirely separate that can spin a container up and down on demand.
Click to expand...
Yip there's no getting away from it. Best to just get them a fck-around box to roleplay their dev fantasies on. When there's 50 apps on that thing each running their own container and 4 concurrent connections, it would be amusing to watch them start blaming each other...
 
You must log in or register to reply here.
Back
Top
Sign up to the MyBroadband newsletter
X