Chat GPT & Generative AI tools

A dedicated space to explore and discuss how cutting-edge technologies are reshaping sports trading and betting strategies. Collaborate on innovative ideas, and stay ahead of the curve in a rapidly evolving landscape.
Post Reply
User avatar
ShaunWhite
Posts: 10424
Joined: Sat Sep 03, 2016 3:42 am

napshnap wrote:
Fri Mar 07, 2025 1:52 pm
- Beep-bop-beep, shut up and keep mining, meat bags!"
This idea it'll try to dominate and take over makes the assumption it will have the same flaws as it's creators. With no ego and no desire to 'progress' then it's conclusion might be quite benign. Or we get melted down to use as robot lube so who knows eh.
User avatar
ShaunWhite
Posts: 10424
Joined: Sat Sep 03, 2016 3:42 am

Bit off topic but seems the right place.

In the words of GPT:
Upgrading to Office 2024 isn’t a simple install—it forces a full uninstall of Office 2019 first, meaning you lose settings, macros, templates, and UI customizations unless manually backed up. Export your ribbon/toolbars, macros (PERSONAL.XLSB), templates (Normal.dotm), and registry settings before uninstalling. Expect some manual fixes post-install. Microsoft isn’t making this hard by accident—they want perpetual license users to suffer just enough to consider switching to Microsoft 365. Also, if you have Visio 2019, it won’t work alongside Office 2024, forcing an upgrade to Visio 2024 or removal.

Luckily I got an MSDN licence for Visio Pro 2024 for £12 (rrp £640!!). They rarely blacklist them and at £12 they could do that 50 times and it would still be cheaper. My last key worked for 6yrs before the uninstall.
sionascaig
Posts: 1620
Joined: Fri Nov 20, 2015 9:38 am

Starting to see some articles on "jailbreaking" various AI's so no longer constrained by ethical or other restrictions / guidelines built in by the developers.

One article also hinted that a jailbreaked bot could then be released back to general public that followed your own instructions / guidelines.

Scary world indeed.
User avatar
Big Bad Barney
Posts: 330
Joined: Mon Feb 04, 2019 6:00 am

sionascaig wrote:
Tue Mar 18, 2025 6:56 am
Starting to see some articles on "jailbreaking" various AI's so no longer constrained by ethical or other restrictions / guidelines built in by the developers.

One article also hinted that a jailbreaked bot could then be released back to general public that followed your own instructions / guidelines.

Scary world indeed.
Not sure they really work like that? You can try GPT4All to run your own models.... they chug unless you got a mammoth video card...

I got a pretty beefy machine (top of the line 2 years ago specced out sorta thing), and they chug.....
User avatar
Euler
Posts: 26317
Joined: Wed Nov 10, 2010 1:39 pm

Trying to trick and get around restrictions is an ongoing game it seems.
You do not have the required permissions to view the files attached to this post.
Fugazi
Posts: 933
Joined: Wed Jan 10, 2024 7:20 pm

Naryank wrote:
Tue Mar 25, 2025 4:46 pm
I've been using AI tools like ChatGPT for a while now, and they're handy for brainstorming ideas, writing drafts, and even troubleshooting code. It saves a ton of time, especially when you're stuck on something and need a quick explanation or example. That said, there's still a lot of room for improvement, especially when handling more complex tasks without human intervention.

For businesses, AI isn't just about chatbots. It's also transforming customer service through voice automation. I came across this AI voicebot solution that helps companies automate support, take orders, and even schedule appointments, saving time and resources. It might be worth a look if you're interested in AI-driven business tools like URL REMOVED
Spam.
User avatar
ShaunWhite
Posts: 10424
Joined: Sat Sep 03, 2016 3:42 am

Getting annoyed with how many issues I'm having to raise with OpenAI and how they take so long to reply. Mostly about ChatGPT not following it's own directives, such as repeated 'promises' to always validate and double check (aka check) output, then immediately using assumptions. Ditto profile directives such as always use UK prices, then always been told USD prices.

And the huge downgrade in validation during peak periods to relieve server stress. Frankly it's been useless.

.........
Key Issues That Remain Unanswered:

ChatGPT has no true memory within a session, but why does it inconsistently follow instructions within the same conversation?
If the model explicitly agrees to follow validation steps but then fails to do so moments later, that suggests a systemic issue rather than just a design limitation.
If this is expected behavior, can you clarify whether OpenAI considers this a known trade-off or an oversight?

"It’s a design limitation, not a bug"—but is OpenAI actively working on improving this?
The fact that ChatGPT sometimes does follow instructions properly but then fails at other times suggests that this isn’t purely a fundamental AI limitation—it’s a consistency issue.
Is OpenAI planning any improvements to instruction adherence, or is this considered an intentional model trade-off?
If the model is prioritizing speed over reasoning depth, is there an option to adjust this?

You mentioned that ChatGPT may take shortcuts to optimize efficiency.
Is this purely an internal model tuning decision, or will OpenAI allow users to adjust response depth vs. speed (e.g., a “strict validation” mode in future releases)?

Reinforcing validation in every prompt shouldn’t be necessary when a directive is in place.
If a directive is set at the profile level, why does ChatGPT require reinforcement in each prompt?
Shouldn’t the model be designed to respect the user’s explicit instructions without constant manual prompting?

Does system load impact response quality?
You suggested testing at different times for better reasoning accuracy.
Are you confirming that model accuracy degrades under high load? If so, what changes are being made to address this?

Final Thoughts & Next Steps
I understand that ChatGPT has limitations, but I’d like a clearer answer on whether OpenAI is actively working on improving instruction adherence or if this is considered an acceptable trade-off in the model’s current form
.
User avatar
Big Bad Barney
Posts: 330
Joined: Mon Feb 04, 2019 6:00 am

ShaunWhite wrote:
Tue Mar 25, 2025 6:44 pm
many issues
Word!.....and like...can it stop talking to me it flattering tones as it's default... Great idea!
Fugazi
Posts: 933
Joined: Wed Jan 10, 2024 7:20 pm

My biggest mistake is going In circles for hours with it trying to solve the same problem, instead of accepting its limitations and moving on. I haven't learned no matter how many times I tell it to stop attempting the same thing it will ignore me and do it anyway 😂
User avatar
ShaunWhite
Posts: 10424
Joined: Sat Sep 03, 2016 3:42 am

Fugazi wrote:
Wed Mar 26, 2025 10:09 am
My biggest mistake is going In circles for hours with it trying to solve the same problem, instead of accepting its limitations and moving on. I haven't learned no matter how many times I tell it to stop attempting the same thing it will ignore me and do it anyway 😂
Yep that's the issue I raised.

Ask it to summarise the thread so far, and start a new thread with that (editted) to reset the logic. And apparently the thumbs up/down responses are more 'sticky' than text comments about it being right or wrong. Does that work, not really but that's the guidance.
makkad
Posts: 102
Joined: Thu Sep 22, 2016 10:38 pm

ShaunWhite wrote:
Tue Mar 18, 2025 2:35 am
Bit off topic but seems the right place.

In the words of GPT:
Upgrading to Office 2024 isn’t a simple install—it forces a full uninstall of Office 2019 first, meaning you lose settings, macros, templates, and UI customizations unless manually backed up. Export your ribbon/toolbars, macros (PERSONAL.XLSB), templates (Normal.dotm), and registry settings before uninstalling. Expect some manual fixes post-install. Microsoft isn’t making this hard by accident—they want perpetual license users to suffer just enough to consider switching to Microsoft 365. Also, if you have Visio 2019, it won’t work alongside Office 2024, forcing an upgrade to Visio 2024 or removal.

Luckily I got an MSDN licence for Visio Pro 2024 for £12 (rrp £640!!). They rarely blacklist them and at £12 they could do that 50 times and it would still be cheaper. My last key worked for 6yrs before the uninstall.
Thanks for this info

I'm upgrading to a Windows11 laptop (current laptop HP8460p won't support W11). My plan was to purchase an Office 2024 licence online (£30-ish?) and install on the new laptop. I wasn't aware of the hurdles one has to overcome. If you have any other info on the pitfalls please post on here. Thx Again
foxwood
Posts: 429
Joined: Mon Jul 23, 2012 2:54 pm

ShaunWhite wrote:
Wed Mar 26, 2025 10:58 am
Ask it to summarise the thread so far, and start a new thread with that (editted) to reset the logic. And apparently the thumbs up/down responses are more 'sticky' than text comments about it being right or wrong. Does that work, not really but that's the guidance.
Trouble is that it now seems to work in "consultant" mode - it'll list a dozen different variations of possible answers without identifying the best in the light of the thread so far.

I've found that it's better when you add to every post that the answer should be brief, without explanation, one-line, concise etc - anything to force a short answer. The thread then tends to stay on topic much longer and gets better results. A summary and new thread also works better since there is less confusion in the content.

Abacus warns when the thread is getting a bit long - almost seems to be word count based which is possibly a sensible measure of confusion anyway.
User avatar
Euler
Posts: 26317
Joined: Wed Nov 10, 2010 1:39 pm

As there is a lot of activity in this area I've created a new forum section for it. So we can discuss specific aspects and share stuff.
bdamir
Posts: 7
Joined: Sun Mar 20, 2022 2:13 am

I’ve used ChatGPT for various tasks, especially for brainstorming and drafting content. It’s impressive how quickly it generates responses, but I’ve noticed it can sometimes be repetitive or lack a natural flow. When using it for coding, it’s helpful, but you definitely need to review and tweak the output, as small errors can slip through.
User avatar
firlandsfarm
Posts: 3318
Joined: Sat May 03, 2014 8:20 am

I have two 'long-term' chats ongoing with GPT, one is to create a homepage for a website and the other to put forward logo/banner designs. Both are faltering at providing some graphics. It keeps telling me it has everything ready but no matter how it tries to upload them for me to retrieve, it fails. When I point out the failure it apologises and says 'I'll do it differently and this time it will work' ... but so far it hasn't.

The homepage task was going great, content was agreed, layout agreed and the HTML file loaded and displayed perfectly but as soon as I asked for some graphics to 'spice it up a little' the download kept failing. Obviously the second task is all about graphics so that makes me think the problem is graphics related.

Is there something I need to do to open a path for graphics? Maybe something in the firewall (standard Windows 11 Pro)? In another chat it happily makes csv files available and I download them without trouble.

'We' have tried direct upload/download, emailing as attachments, uploading to a shared onedrive folder and more. Any thoughts/suggestions would be welcome please ... this whole project has stalled because of this. :(
Post Reply

Return to “AI, Machine Learning and Generative AI”