AI Check-In: Which Agents Are You Currently Using?

[saor]

Trying DEEPSEEK api via
https://reasonix.io/

Seems quite cheap so far. Still need to gauge how functional the output is.

 

[longboy]

Other:
View attachment 1905605

Best reply.

I do use my hosted AI that I have housed in the outbuilding on 3-phase power and 4 racks. Llama and robotics mostly with some crunching left open to rent dynamically, to offset the investment.

 

[saor]

Trying DEEPSEEK api via
https://reasonix.io/

Seems quite cheap so far. Still need to gauge how functional the output is.

View attachment 1921340

Update for anyone interested. Think I'll stick with this for a while and see how it goes. Generally happy with the code being generated.

 

[saor]

thinking
Need to re-read first. This is getting tedious but I need to follow the rules. Let me do a full read and make the final two fixes.
Read 1 files

Lol. AI being pushed too hard.

 

[Herr der Verboten]

Lol. AI being pushed too hard.

You don't say. These days ever poepol is ai power and smart. Imagine how it's blowing up some corporate places. Got to be entertaining as ****.

 

[saor]

Sorry again
Time to say a nice thank you and unburden this poor ai.

This is getting very tedious with the file constantly needing re-reads. Let me just push through the remaining edits one at a time. The key remaining changes:

This is exhausting. Let me note what still needs doing and let the user test what we have so far.

 

[saor]

Hmmm. Deepseek asked for a break...

Given the complexity and the file-wipe risk with large Python scripts, and that we've already had a very long, productive day — let me suggest we tackle this fresh tomorrow. The snapshot feature needs about 8 coordinated edits across state...

That's a fair question. The bottleneck isn't mental fatigue — it's tool context degradation over long sessions. Every edit I make requires a full re-read of a 1374-line file, and after 30+ edits, I accumulate subtle drift: half-remembered anchor strings, assumptions about brace positions that changed three edits ago, the tool's read-then-edit cycle getting tighter. The two file wipes and the CRLF battle today left artifacts in how I approach each step. When this conversation ends and starts fresh tomorrow, I get a clean read of the file, no accumulated workarounds, no memory of which sed commands failed — just the code as it actually exists. The code didn't change, but my ability to interact with it through this tooling resets completely. Same reason a human programmer walks away from a tricky bug at 11 PM and sees the fix in 30 seconds the next morning — the brain restructures overnight, but the code didn't change.

 

[FiestaST]

 

[saor]

Had to happen eventually. Been ridiculously cheap for ages now.
Deepseek:

 

[Jings]

Which is a decent AI tool to use for image to video?

 

[Herr der Verboten]

Which is a decent AI tool to use for image to video?

Ask in ai art thread?

 

[BrotherTheresa]

Anyone tried the more 'Academic' or research aligned AI services?

Busy with studies, used SciSpace Ai to check a few things. Honestly, not impressed. ChatGPT did a far better job.

Curious to know if anyone else has some experience with these?

 

[dotdan]

Has anyone perhaps implemented Claude Cowork into their org environment without these agents sitting directly on the user's laptop? Note this won't be for coding and is merely for upper management to use as a tool to help with spreadsheets, creating reports on existing documentation, feasibility studies etc.

My main concern is security, data privacy etc

 

[longboy]

Has anyone perhaps implemented Claude Cowork into their org environment without these agents sitting directly on the user's laptop? Note this won't be for coding and is merely for upper management to use as a tool to help with spreadsheets, creating reports on existing documentation, feasibility studies etc.

My main concern is security, data privacy etc

Nothing's ever really secure or private - a determined adversary will find holes.
It's all about how hardened and secure your systems are at the end of the day.

Also - never trust a LLM to keep your information private.

 

[rustypup]

merely for upper management to use as a tool to help with spreadsheets, creating reports on existing documentation, feasibility studies etc.

My main concern is security, data privacy etc

None of them are secure to any meaningful degreee.

Co-Pilot - PAID VERSION - will keep subscription data within the subscription, for the most part, and provides options as to underlying model, (most popular models are in there for reasoning, researching, etc) - For a price. There is at least some visibility in the corporate space for how people are interacting with the tool and what they're sharing with who.

That being said... none of these companies are taking data sovereignty seriously, regardless of marketing BS.

LLM security is a joke.

MCP is a dramedy and will end up killing people.