All things AI

[PotNoodleTech]

"The government says the assistants - called Humphrey - will "speed up the work of civil servants" and save money by replacing cash that would have been spent on consultants."

 

This is the main problem - why they heck do they constantly need consultants? Do they not have a clue about their job? Are they not willing to learn new things? Is there no training courses for them?

[ITGuyNW]

Anyone had any issues with PowerPoint/Office and an error saying ai.exe bad image ?

 

Seems Office has had an update recently to its AI functionality. Had to put some exclusions in to Sophos and Impero to get rid of the error.

[CHiLL]

Anyone had any issues with PowerPoint/Office and an error saying ai.exe bad image ?

 

Seems Office has had an update recently to its AI functionality. Had to put some exclusions in to Sophos and Impero to get rid of the error.

We had a report from a user of this once last week but it never while we had the laptop and we've not heard anything since and haven't added any exclusions or made any changes for this

[SDM01]

I would be interest in hear what schools are using as AI content detectors for plagiarism that integrates with Teams. I've looked at solutions like Turnitin but really need to try and find a more cost effective solution while not adding to the teachers workload.

[CHiLL]

Anyone had any issues with PowerPoint/Office and an error saying ai.exe bad image ?

 

Seems Office has had an update recently to its AI functionality. Had to put some exclusions in to Sophos and Impero to get rid of the error.

Could you clarify where you entered your exceptions? I presume you mean in your AV settings?

[ITGuyNW]

Could you clarify where you entered your exceptions? I presume you mean in your AV settings?

 

Yeah we used the global exclusions setting in Sophos Central and we host Impero Ed Pro on prem, so in server go to tools > injection exclusions. Might take a client reboot to recognise the changes.

[Ditto]

Amazing! Not a single post on DeepSeek - well not until now

[jthompson]

Amazing! Not a single post on DeepSeek - well not until now

Investors Worried AI Could Leave Shred Of Humanity Intact

 

https://theonion.com/investors-worried-ai-could-leave-shred-of-humanity-inta-1850257921/

[Theblacksheep]

OpenAI concerned that Deepseek has used their IP to make their product, is ironic considering OpenAI has more stolen content than the PirateBay.

[dmj]

OpenAI concerned that Deepseek has used their IP to make their product, is ironic considering OpenAI has more stolen content than the PirateBay.

 

Another irony is that the US restricted China from importing nvidia to stifle their innovation, and this resulted in them creating a more efficient technology

[6Foot2]

The Guardian tests AI:

 

ChatGPT

Claude

DeepSeek

Gemini

Grok

Meta AI

 

Link: DeepSeek, ChatGPT, Grok … which is the best AI assistant? We put them to the test | Artificial intelligence (AI) | The Guardian

 

Spoiler: They don't declare a "winner"

[6Foot2]

 

Thanks for clearing that up!

 

[Andrew_C]

Seeing reports that DeepSeek has been hit with 230m DDOS requests per second. Do you think they've got up someone's nose?

[6Foot2]

Seeing reports that DeepSeek has been hit with 230m DDOS requests per second. Do you think they've got up someone's nose?

 

The report of the attack may have been a "joke":

 

Link:

 

[6Foot2]

Link: Google remakes Super Bowl ad after AI cheese gaffe - BBC News

 

Google has re-edited an advert for its leading artificial intelligence (AI) tool, Gemini, after it overestimated the global appetite for Gouda.

 

The commercial - which was supposed to showcase Gemini's abilities - was created to be broadcast during the Super Bowl.

 

It showed the tool helping a cheesemonger in Wisconsin write a product description by informing him Gouda accounts for "50 to 60 percent of global cheese consumption".

 

However, a blogger pointed out on X that the stat was "unequivocally false" as the Dutch cheese was nowhere near that popular.

 

Replying to him, external, Google executive Jerry Dischler, insisted this was not a "hallucination" - where AI systems invent untrue information - blaming the websites Gemini had scraped the information from instead.

 

"Gemini is grounded in the Web - and users can always check the results and references," he wrote.

 

"In this case, multiple sites across the web include the 50-60% stat."

 

The ad has now been re-edited to remove the error.

 

Google has posted it on YouTube, external - which it owns - and it no longer contains any reference to what percentage of the world consumes the hard cheese.

 

In a statement, Google told the BBC it had remade the ad after speaking to the cheesemonger featured in it and asking him what he would have done.

 

"Following his suggestion to have Gemini rewrite the product description without the stat, we updated the user interface to reflect what the business would do," the statement added.

 

Accurate figures on cheese are hard to pin down - however it is thought that cheddar is in fact the most popular variety in the world, though preferences vary from country to country. Another major contender is mozzarella, external, given its use in pasta and pizza.

 

AI vs cheese

 

The incident is embarrassing for Google, given the extremely high profile of the Super Bowl and the extra scrutiny ads for it receive.

 

Blogger Nate Hake - who spotted the error, external - said it was an example of "AI slop".

 

But it's not the first time the tech giant has found itself on the back foot over its AI products.

 

A year ago, Gemini was "paused" following criticism of the "woke" images it generated, such as an image of the US Founding Fathers which inaccurately included a black man.

 

It's even had previous issues with cheese.

 

In May last year, its AI overviews search feature was criticised after it told some users to use "non-toxic glue" when they searched for 'how to make cheese stick to pizza better'.

 

The search engine's AI-generated responses also said geologists recommend humans eat one rock per day.

 

Problems with AI tools are not limited to Google - in January, Apple was forced to suspend its news alert summariser after it invented a slew of inaccurate headlines.

 

Adverts at the Super Bowl themselves are also no stranger to controversy.

 

Last year, Uber Eats made a last minute change to its advertisement after criticism that it inappropriately made light of food allergies.

[Ditto]

I suggest Jerry Dischler was right about it not being a 'hallucination' - but it could have been a 'halloumination'

[Rob_D]

https://www.forbes.com/sites/danfitzpatrick/2025/02/03/new-copyright-ruling-just-made-ai-skills-the-biggest-advantage/

The U.S. Copyright Office has concluded that AI-generated work can be copyrighted when it embodies meaningful human authorship. This is big news. It means that those who develop expertise in working with AI can be rewarded with intellectual property.

 

US seems to be doing similar things to the UK when it comes to AI copyright. I'm not sure if it makes that much difference in the grand scheme of things, but it feels like it could.

[Ditto]

Following a post in the BETT thread, with a view to AI use in products, and also a thread about timetabling it got me thinking. ChatGPT has impressed me a lot recently, and I'm initially sceptical of the AI bandwagon. Some of that maybe from a reasonably deep understanding of Text Analytics when I led a team at IBM - it's a challenging topic, but advances in the last decade are staggering. To bring the topic threads together, I'm starting to wonder, by providing all of the input factors that go to pulling a timetable together, like classrooms, teacher subjects, working times, pupil numbers per subject etc. etc., could you see an AI putting together full TT options based on that input, offering several TT options. I think it is entirely achievable - what do others think?

[Aftab_Ahmed]

Given all the discussions on AI within MIS systems, NSPCC have released this interesting report in the safety aspects of using AI which I thought would share with you all.

 

Happy reading on this report

NSPCC generative-ai-childrens-safety.pdf

[6Foot2]

[browolf]

I built ourselves an ai cluster on old servers. Partly as a challenge and also so ICT can teach ai without worrying that platforms change or limit access ruining their lessons. It uses Open web ui and nginx as a reverse proxy to 3 servers running llama3.2:1b on ollama. Edited February 12, 2025 by browolf

[PotNoodleTech]

I built ourselves an ai cluster on old servers. Partly as a challenge and also so ICT can teach ai without worrying that platforms change or limit access ruining their lessons. It uses Open web ui and nginx as a reverse proxy to 3 servers running llama3.2:1b on ollama.

 

That's a cool idea actually, secure internal AI to play with. Did you use a GPU or using all CPU?

[browolf]

That's a cool idea actually, secure internal AI to play with. Did you use a GPU or using all CPU?

 

1 dual processor hpe server with 240Gb ram, the old datto server with 64Gb ram. A xeon workstation with 64Gb ram & an nvidia GPU. Ollama installs for whatever is available but the xeon workstation is fastest on responses because of the GPU.

Edited February 12, 2025 by browolf

[Ditto]

I presume there is a LLM involved somewhere - where is that resource hwls? Do you have to have access to an external resource for that?

[SteveM555]

I presume there is a LLM involved somewhere - where is that resource hwls? Do you have to have access to an external resource for that?

 

Ollama is the engine that runs the LLM models, so it's all local. The models can be pulled directly from Ollama (deepseek r1, llama3, mistral, gemma etc.) with varying parameter size so it can scale on lower-end hardware if needed.

 

Once the models are local it doesn't need any external resources, or even internet access. OpenWebUI provides a frontend for communicating with the models which is very similar to ChatGPT. We've been playing around with it and have LDAP logins working, monitoring and running a few 7b / 2b models. It works really well.