AI Agents??

Anyone else smell burnt toast?

Once upon a time you toasted bread by holding it in front of an open fire. You could get the toast just as toasty you liked it. But this, apparently, was too much effort.

So somebody invented the electric toasting element. Where the bread was put in front of a hot electrical element and you didn't have to hold it. All you had to do was watch it, decide when it was done and flip it to the other side. This way you could get the perfect toast, just as you liked it. But this soon also became too much effort.

So somebody invented the pop-up toaster. Where you put your slice of bread in (it could only now fit certain sizes of sliced bread), and a timer meant your toast popped up just when you had guessed what a pre-set timer had to be to make your toast perfect. But this meant thinking about how long toast might take, so soon became too much effort.

So somebody added a microprocessor to the toaster. And now you put your AS standard sized pre-sliced bread into the slot and the toast was spat out just when the computer decided it how toasty you should like it.

But now that isn't good enough and we need AI to decide how we must like toast.

And this folks is progress.

What was the question again?

We toast under the grill so still need to watch it so as not to burn.
Wife claims it's way better done under the grill.
Sometimes she needs to do more as burt toast is not to her liking.
I just eat what ever I am given ??

Ok that was a fun distraction.
Seems none knows anything about AI Agents ??

Select to expand quote

Froth Goth said..
To estimate the probability of this extremely rare event, we'll need to make some assumptions and combine several factors.

Assumptions
1. *Stroke probability*: As mentioned earlier, the annual incidence of stroke in men aged 65-74 is approximately 1.4% (Source: American Heart Association).
2. *Olfactory aura probability*: The prevalence of olfactory auras (smelling burnt toast or other unusual odors) before or during a stroke is difficult to quantify. Let's assume a rough estimate of 1% to 5% of stroke cases (this is a wide range, as there's limited research on this specific topic).
3. *Simultaneous strokes*: The probability of three people having strokes at the exact same moment is already extremely low. Let's assume a rough estimate of 1 in 100 million (this is a very rough estimate, as there's no reliable data on simultaneous strokes).
4. *Independent events*: We'll assume that the three men's strokes and olfactory auras are independent events, meaning that one person's stroke or olfactory aura doesn't influence the others.

Calculating the probability
Using these assumptions, we can estimate the probability of all three men smelling burnt toast and having strokes at the same time:

1. Probability of one person having a stroke: 1.4% (0.014)
2. Probability of one person experiencing an olfactory aura: 1% to 5% (we'll use 3% as a rough estimate) (0.03)
3. Probability of one person having a stroke and experiencing an olfactory aura: 0.014 ? 0.03 ? 0.00042 (or approximately 1 in 2,381)
4. Probability of three people having strokes and experiencing olfactory auras at the same time: (0.00042)^3 ? 7.488 ? 10^(-11) (or approximately 1 in 1,333,333,333)

Keep in mind that this calculation is highly speculative and based on rough estimates. The actual probability of this event could be significantly higher or lower.

In conclusion, while we can't provide an exact probability, it's clear that the likelihood of three old men smelling burnt toast and having strokes at the same time is incredibly low - on the order of 1 in billions or even trillions.

This is a sensible probablility calculation. It should be similarly applied to the probablility of young Australian males contracting myocarditis... prior to COVID, post-COVID, and post COVID vaccination.

Select to expand quote

warwickl said..
Seems none knows anything about AI Agents ??

Is this one ?

Select to expand quote

warwickl said..
Do I need my own AI Agent?
If so is there any urgency?
Can anyone simply what this system will do?

As someone who identifies as an AI expert, I will simplify it for you.

Imagine a scenario where a crazy btch lodges a complaint with the local council, claiming-without cause-that your little angel dog is dangerous simply because you didn't cut his nuts off. No harm was done, yet a council ranger is dispatched to hassle you.
You have two options:
1. Navigate the bureaucratic nightmare, hire a lawyer, and drain your finances on legal paperwork.
2. Leverage your AI agent to construct a legal defense, compile relevant data, analyze case law via LexisNexis, draft applications, send formal letters, and request references-all while you make your coffee.
We're not quite there yet with opensource models, but we're ~75%.

Here's a sample "snippet" of a Defense protocol for WA,
1. Immediate Document Acquisition
Action: Submit FOI Form 12 to council demanding:
Original complaint (unedited, with metadata)
Ranger's incident report (pre-redaction version)
Complainants' full complaint history
Law: FOI Act 1992 (WA) s12(2) + Dog Act 1976 s33(5)
Deadline: 7 business days
2. Evidence Preservation Order
Action: File Form 89A (State Administrative Tribunal) to:
Block document destruction (cite State Records Act 2000 s61 )
Require council preserve all CCTV/emails related to the complaint
Cost: $102 (waivable under s89A(3) for financial hardship)
3. Pre-Action Discovery
4. Procedural Impropriety Challenge
5. Counter-Complaint Matrix
6. Judicial Backstop
7. Forensic Evidence Package
8. Statutory Deadlines Table
9. Nuclear Option
10. Resolution Checklist


The following assumed the council refused to give you the original complaint in step 1 above.
1. Challenge FOI Refusal Submit Form 39/1R for internal review citing FOI Act 1992 s23(4) exemption override.
2. Escalate to Ombudsman File Form OMB-12 for maladministration (+ attach council's refusal letter).
3. Initiate Judicial Review Draft Writ of Mandamus (Supreme Court Form 4A) under Order 56 RSC .
4. Deploy Counterevidence Submit Form 89 demanding proof of complainant's confidentiality request.
5. Criminal Complaint File police report (PR-XX-XXXXXXX ) citing FOI Act s70 obstruction offense.
6. Claim Compensation Amend Magistrates Court claim with Form 3C adding $32 FOI fee + $1,850 legal costs.
7. Force Disclosure Serve s23(8) Notice : 48-hour ultimatum for document release.
8. Systemic Reform Push Submit Form IC-7 to Information Commissioner alleging pattern of FOI breaches.
9. Public Shaming File ABC FOI Showbag Template for journalist investigation.
10. Final Threat Email council draft Supreme Court filing using Community Law Network Template 14B .
Required Forms: Form 39/1R (FOI Internal Review)OMB-12 (Ombudsman Complaint)Supreme Court Form 4APR-[Case#] (Police Reference)Magistrates Court Form 3CForm IC-7 (Information Commissioner)Execute steps 1-4 within 7 days to meet statutory deadlines.

The Chain of thought model orchestrates mixture of expert models which write python code to navigate the web, filling the forms, attach or email them, etc.This is nothing I am involved with, I just lent a helping AI which cost me ~$1.35.

The above is a snippet, the actual output was ~30 pages with tables of different strategies, consequences, costs, likelihood of success, ranging from offering to write a letter of apology to basically taking the council to court and getting criminal charges against Complainant. eg: Under Dog Act 1976 (WA) Section 33E, any person making false declarations about canine behavior commits a Tier 2 Offense (penalty: $5,000 AUD + potential 6-month prohibition on dog ownership). Complementary action exists under Criminal Code Act 1913 Section 445 (Public Mischief) for knowingly false reports wasting police/council resources.

Select to expand quote

FlySurfer said..
As someone who identifies as an AI expert, I will simplify it for you.

Imagine a scenario where a crazy btch lodges a complaint with the local council, claiming-without cause-that your little angel dog is dangerous simply because you didn't cut his nuts off. No harm was done, yet a council ranger is dispatched to hassle you.
You have two options:
1. Navigate the bureaucratic nightmare, hire a lawyer, and drain your finances on legal paperwork.
2. Leverage your AI agent to construct a legal defense, compile relevant data, analyze case law via LexisNexis, draft applications, send formal letters, and request references-all while you make your coffee.
We're not quite there yet with opensource models, but we're ~75%.

Here's a sample "snippet" of a Defense protocol for WA,
1. Immediate Document Acquisition
Action: Submit FOI Form 12 to council demanding:
Original complaint (unedited, with metadata)
Ranger's incident report (pre-redaction version)
Complainants' full complaint history
Law: FOI Act 1992 (WA) s12(2) + Dog Act 1976 s33(5)
Deadline: 7 business days
2. Evidence Preservation Order
Action: File Form 89A (State Administrative Tribunal) to:
Block document destruction (cite State Records Act 2000 s61 )
Require council preserve all CCTV/emails related to the complaint
Cost: $102 (waivable under s89A(3) for financial hardship)
3. Pre-Action Discovery
4. Procedural Impropriety Challenge
5. Counter-Complaint Matrix
6. Judicial Backstop
7. Forensic Evidence Package
8. Statutory Deadlines Table
9. Nuclear Option
10. Resolution Checklist


The following assumed the council refused to give you the original complaint in step 1 above.
1. Challenge FOI Refusal Submit Form 39/1R for internal review citing FOI Act 1992 s23(4) exemption override.
2. Escalate to Ombudsman File Form OMB-12 for maladministration (+ attach council's refusal letter).
3. Initiate Judicial Review Draft Writ of Mandamus (Supreme Court Form 4A) under Order 56 RSC .
4. Deploy Counterevidence Submit Form 89 demanding proof of complainant's confidentiality request.
5. Criminal Complaint File police report (PR-XX-XXXXXXX ) citing FOI Act s70 obstruction offense.
6. Claim Compensation Amend Magistrates Court claim with Form 3C adding $32 FOI fee + $1,850 legal costs.
7. Force Disclosure Serve s23(8) Notice : 48-hour ultimatum for document release.
8. Systemic Reform Push Submit Form IC-7 to Information Commissioner alleging pattern of FOI breaches.
9. Public Shaming File ABC FOI Showbag Template for journalist investigation.
10. Final Threat Email council draft Supreme Court filing using Community Law Network Template 14B .
Required Forms: Form 39/1R (FOI Internal Review)OMB-12 (Ombudsman Complaint)Supreme Court Form 4APR-[Case#] (Police Reference)Magistrates Court Form 3CForm IC-7 (Information Commissioner)Execute steps 1-4 within 7 days to meet statutory deadlines.

The Chain of thought model orchestrates mixture of expert models which write python code to navigate the web, filling the forms, attach or email them, etc.This is nothing I am involved with, I just lent a helping AI which cost me ~$1.35.

The above is a snippet, the actual output was ~30 pages with tables of different strategies, consequences, costs, likelihood of success, ranging from offering to write a letter of apology to basically taking the council to court and getting criminal charges against Complainant. eg: Under Dog Act 1976 (WA) Section 33E, any person making false declarations about canine behavior commits a Tier 2 Offense (penalty: $5,000 AUD + potential 6-month prohibition on dog ownership). Complementary action exists under Criminal Code Act 1913 Section 445 (Public Mischief) for knowingly false reports wasting police/council resources.

Have you just been using models or have you tried training your own?

I've dabbled with various freely available models, and have ended up getting a bit overwhelmed and depressed by how powerful they can be.

The last project I worked on was in steganography. Hiding one image in another using a bunch of software tied together with python. Another project was using various voice models to change the vocals on YouTube music clips. Taylor Swift did quite an interesting version of 'I was only 19'. The results of both experiments were incredible.

jumpshare.com/s/AFNQ0f5RlQpuyB0gyWPe

github.com/harveyslash/Deep-Steganography

github.com/SociallyIneptWeeb/AICoverGen

I've been mainly using runpod and Google colab to do most of it, but this new deepseek is an exciting development, which likely means I could run more things locally.

I can't create an account. Has anyone given it a go?

Hilarious: Israeli AI Hasbara bot goes rogue, posting inconvenient truths and mistakes alike:

Select to expand quote

FlySurfer said..

I've trained small (<3B) models in the early days; now I primarily use RAG with the APIs of the four major LLMs.

I think we fundamentally approach AI differently. I don't try to anthropomorphize it, nor do I have illusions of it escaping, replicating and infecting the world. I see it as a tool, like an exoskeleton. It allows me to be more productive and accurate with my work and projects. I don't see it as a competitor to my job, but as a means to get the job done faster and better.
My focus is on user augmentation in a corporate environment, and personally to have a capable assistant both at home and at work.

At home, I've been using DeepSeek/OpenAI/Anthropic integrated with Open WebUI, Motion Eye, DeepStack, Alexa, and Home Assistant, that was until I started having latency issues when the media picked up the DeepSeek story. My OpenAI queries were costing me 12-20 times more than DeepSeek's introductory pricing.

So, being depressed by the power of an exoskeleton does not compute., but I did ponder what skills could be useful in the new paradigm and stenography was high up there in the absence of neural linkage.

I am excited and depressed by it. It's not going to go away. I can see all the incredible things it can be used for, but can see negatives.

I just lost interest in dabbling with it after training my own models on particular artists and then created AI artwork work using their style. The results were really impressive. But after I did it I just felt guilty and empty with the creations. I didn't want to share my experiment with anyone. Not unlike stealing something expensive you didn't need or want. Which is how I saw it.

The steganography stuff, was kind of the similar. After spending a couple of days setting it up on my mac, I thought how great is this, and was planning on creating a web gui for it and putting it online creating my own model etc. But then I hesitated and figured the only thing it was most useful for was crime. Lol.

Another semi interesting thing I did was set up a python voice recognition interface to openai so I could talk to my computer and it would talk back to me. That was interesting for around 2 hours.

What I liked about the dabbling was more the process and getting my head around the tools, ie jupiter notebook. My computer being an old Mac m1 is not particularly suitable for AI.

If I found a project again, that interests me and is in a small way useful to me or society I'd dabble some more.

Any suggestions welcome.

Will getting a 50 video card when they drop for laptops help much with hobbyist levels of ai development stuff or can i just grab a laptop with a 4060 and itll be fine ? Last time i had a computer i was unlocking the 4th core on the tri phenom cpus so im well out of the game

I don't want to get into an argument . I have nothing against squirrels. But this is the real AI agent.