Chatgpt - Seriously good potential (or just some Internet fun)

GPT5 honestly feels like shrinkflation.

Getting frustrated with GPT5. I've being using GPT to keep track of food intake but since going to 5 it has been all over the place. I then tried the Google one on my phone and it told me to call a mental health charity after it suggested I was getting too much protein and I questioned it.

Yes, 5 has forgotten a chunk of some formatting rules I've been using, which took a little while to fix, but it's still better than doing it by hand.

They have made a total fudge of the release of this "new" GPT5, as evidenced by the frantic backpedalling they are doing on X.

After all the hyping up they did it's basically a model switcher which has properly screwed up the way the existing products worked.
Forgetting stuff suddenly mid conversation, hot and cold censorship, split personality syndrome, miniscule levels of context retention.... using GPT5 is like playing the lottery in terms of output quality.
 
To be fair, it has got better now. Can't say I notice any difference from the previous model, whether that is good or not, I'll leave to someone else to decide.
 
I have used ChatGPT a bit to help write/edit a few basic programs for me and I find it to be at the same time very clever and immensely stupid!

For instance I have a fairly simple bash script I use when I do a clean Linux Mint install. ChatGPT has added some nice bells and whistles and the first iteration was actually not bad and just needed a bit of tweaking. Two days later though it's forgotten what it did multiple times, added apt packages that weren't in my original script and removed packages that were, messed up a crontab entry so it errored when the script ran and removed entire sections of system tweaks that I didn't ask it to. It seems every time you ask it to modify one thing it changes three that you didn't ask it to.

Quite infuriating sometimes, but it's still helped me create things in hours that would probably take me weeks with my extremely limited programming skills.
 
I've found GPT 5 to actually be fair bit slower, but there's a good chance i'm asking it more complex things than I did 4o!
GPT5 by default has an auto model selection, so it might be switching to a more complex reasoning model for some of your queries. You can force model selection.
 
tried this
Google's NotebookLM will now transcribe YouTube videos. The resulting text is not formatted at all - just one big block - but you can add it to a notebook and ask questions based on it.

willl AI summarize youtube videos textually or aurally, & give 'mind-maps' -
has worked on those I tried, but maybe previous real-time viewings had already transcipted them.
- so, no need even to sample videos anymore to see if they could be worth watching, or just derivative/plagiarized material.
 
Just don't ask GhatGPT to try edit some photos. Unless you want a good laugh that is...
I recently got married and a few photos were out of focus on the subjects. So out of curiosity I thought "Let's see what ChatGPT can do"

Damn! I've not laughed this hard for some time. It did a great job of the focus but it replaced all the subject faces with some random people. And also seemed to add some proper freaky expressions :cry:

But in regards to it being a serious platform. Nah. It just pulls information from the internet which is mostly BS that people have spurted out.
 
Does any one here have some good prompts and / or configurations you use for it? I dabble with it from time to time, and we use copilot at work which now has gpt5 configured on it and want to get the best out of it.
 
I still think it's a bit of fun. I had a whinge at it earlier.

I6v0o8C.png

But having said that, it's helped me out with basic bash scripting many times.
 
Does any one here have some good prompts and / or configurations you use for it? I dabble with it from time to time, and we use copilot at work which now has gpt5 configured on it and want to get the best out of it.
Prompts to do what?

Everything you do needs it own prompt. A prompt to write a SQL query is going to be completely different to a prompt to extract sentiment from an email or translate a PDF into German.
 
Ironic Zuck takes down false holocaust stories
At first glance, it is a heartwarming photo: a little girl feeding the ducks on the canal in pre-war Amsterdam.
but on the other hand they are planning to use AI for
Abstract
As the number of Holocaust survivors declines, preserving their narratives becomes critical. Articial intelligence (AI)
has emerged as a signicant tool in Holocaust education, enabling immersive experiences, interactive storytelling, and the
preservation of survivor testimonies. This study presents the rst empirical comparison of four generations of Holocaust
descendants’ perceptions of AI-generated Holocaust content. Findings reveal a Generational Trust Gap: while the rst and
second generations prioritize emotional authenticity and remain skeptical of AI, third- and fourth-generation participants
show greater trust in digital platforms and AI-driven tools for Holocaust education. This shift does not reect disengage-
ment, but rather a transformation in learning practices

pity they can't be as robust on moderation of content surrounding Gaza (...or UK political parties ...)
 
Prompts to do what?

Everything you do needs it own prompt. A prompt to write a SQL query is going to be completely different to a prompt to extract sentiment from an email or translate a PDF into German.
You can personalise it under settings and make it behave a certain way/customise it's responses based on user preferences. I just wondered what people choose for their interactions with it.
 
Ironic Zuck takes down false holocaust stories
At first glance, it is a heartwarming photo: a little girl feeding the ducks on the canal in pre-war Amsterdam.
but on the other hand they are planning to use AI for


pity they can't be as robust on moderation of content surrounding Gaza (...or UK political parties ...)

They could provide simple RAG model tokenised access to document and artefacts that have been vetted and validated. They could even provide a cryptographically secure mechanism for verifying the details.

In the end if someone wants to take that information, modify, falsify or utilise in non-representative purposes then it's possible - but what is a trusted site is key as a beacon of truth.

Issue seems to be the nazi right wing, now the survivors are dwindling to maintain the truth, are attempting to subvert and modify history.
 
I’m finding I’m using it at least a couple of times a week, mainly uploading an image of something I might be interested in and asking for reviews, best place to purchase, best price, known issues
 
I think generative AI has a long way to go - however for general writer's block and a quick rough and ready bit of research it can be quite good. I mean at work it's tied into our ecosystem (so it doesn't go out and we can share or point it to a sharepoint site) so it can be useful to extract information and find key contacts, summarise meetings etc... I also sometimes use it to help me do some bits in excel or powerBI when I'm stuck. It effectively speeds up googling for something. I don't feel threatened by it, and it still hallucinates A LOT.
 
Interesting article on what value generative ai has given businesses in the US.
never mind, coming soon to a town near you, we have a 22bn microsoft data centre investment ,
Does sound like mechanism for monopolizing UK market (like Amazon, & offshore profits), or worse an American gift horse, to harvest data, and, run the algorithm/profile customers.

How much updating of existing data centre hardware is occurring to ensure AI implementation efficiency - multi-cored etc.
 
never mind, coming soon to a town near you, we have a 22bn microsoft data centre investment ,
Does sound like mechanism for monopolizing UK market (like Amazon, & offshore profits), or worse an American gift horse, to harvest data, and, run the algorithm/profile customers.

How much updating of existing data centre hardware is occurring to ensure AI implementation efficiency - multi-cored etc.
Do we need to think of a question for that answer?:eek:
 
Back
Top Bottom