Artificial Intelligence thread

[Hyper]

I don't think it matters that closed models like gpt are scoring 1480 vs qwen at 1428. These scores are so full of uncertainty and bias that there will be some variability anyway and if they are western benchmarks then there will be a bias towards English which will put Chinese models at a disadvantage.

Moreover, China doesn't really have to top when they are spending far less on AI compared to the colossal spending made by US companies. I think Chinsse companies are being efficient and that is the most important part. Its important to spend less but still attain a good enough score. Cause sooner or later this AI bubble will burst and funding will dry out.

Then the real battle for AI will start and efficient Chinese companies will benefit the most.

Microsoft is being efficient. They offloaded a ton of spending to CoreWeave that has a bunch of uncertain debt. It outsourced any vulnerability.

 

[HighGround]

I know you're asking someone else but in case you're interested:

I use Gemini 2.5 Pro for to help with clinical work in psychology, such as case analysis, summarizing progress, treatment planning suggestions, integration of different theories, creating meditation scripts and worksheets, etc., I found it performing better in most cases than Claude 4.1, GPT-5, and other Chinese models like Deepseek R1, Qwen 3 etc. The advantage is not necessarily in insight or knowledge (most SOTA AI would cover similar keypoints), but it often has the most polished output, with the right amount of detail (others are sometimes too brief), with the most natural tone and pacing when making verbal scripts, and overall very good instruction following.

I have a personalized tool that let me consult multiple AI at once using the same prompt, so I cross compare them every time I submit a query.

For programming work I still use Claude and Qwen Code.

I work in the legal industry for local tovernment.

Most LLMs are pretty good at generating docs, pdfs, and helping me with excel docs. I draft legal documents, track court stats, and case flow.

IMO, LLMs are invaluable for the service industry in US, and really increase efficiency for people who can use them.

I pay probably ~60$ per month for all my AI crap, but it’s helped me earn roughly ~8$k more in the last year in raises. So the tradeoff has been positive so far.

y’all have to remember that for US, a lot of the GEP here is service based and LLMs can be massive work multipliers in many industries.

 

[Maikeru]

I work in the legal industry for local tovernment.

Most LLMs are pretty good at generating docs, pdfs, and helping me with excel docs. I draft legal documents, track court stats, and case flow.

IMO, LLMs are invaluable for the service industry in US, and really increase efficiency for people who can use them.

I pay probably ~60$ per month for all my AI crap, but it’s helped me earn roughly ~8$k more in the last year in raises. So the tradeoff has been positive so far.

y’all have to remember that for US, a lot of the GEP here is service based and LLMs can be massive work multipliers in many industries.

Seriously, don't use non-specialist LLMs for court work - the documents will be full of hallucinations and outdated or outright made up citations. You need to use proprietary AIs from either law firms or LexisNexis, etc. They have access to paywalled case law databases.

Please, Log in or Register to view URLs content!

On top of which you've got data protection issues if you're putting client data into the LLM.

 

[HighGround]

Seriously, don't use non-specialist LLMs for court work - the documents will be full of hallucinations and outdated or outright made up citations. You need to use proprietary AIs from either law firms or LexisNexis, etc. They have access to paywalled case law databases.

Please, Log in or Register to view URLs content!

WestLaw is useful for researching case work, but this isn't really relevant to my work. 99% of the things I (most people actually) draft have nothing to do with citing cases or statutes.

On top of which you've got data protection issues if you're putting client data into the LLM.

I didn't think about that. But what are the chances this is going to produce a problem?

 

[Maikeru]

WestLaw is useful for researching case work, but this isn't really relevant to my work. 99% of the things I (most people actually) draft have nothing to do with citing cases or statutes.



I didn't think about that. But what are the chances this is going to produce a problem?

Do you want to be the guy to find out?

You can use GPT et al for summarizing cases and drafting letters but always check the output.

 

[siegecrossbow]

WestLaw is useful for researching case work, but this isn't really relevant to my work. 99% of the things I (most people actually) draft have nothing to do with citing cases or statutes.



I didn't think about that. But what are the chances this is going to produce a problem?

Pretty sure you can have a model running locally instead.

 

[HighGround]

Pretty sure you can have a model running locally instead.

Not in the office, I'm lucky my work laptop has more than 8 gigs of RAM!

Do you want to be the guy to find out?

You can use GPT et al for summarizing cases and drafting letters but always check the output.

The output is always checked by many different eyes, that's not really the problem. There's a limited amount of PID being input, specifically emails and case numbers.

I'll have to think about it.

 

[GulfLander]

DeepSeek's launch of new AI model delayed by Huawei chip issues, FT reports​

Please, Log in or Register to view URLs content!

 

[Wangxi]

https://twitter.com/i/web/status/1955856343761592624

https://twitter.com/i/web/status/1955889039116005520