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


An international team of scientists including those from Australia's Charles Darwin University (CDU) has developed a novel AI model known as ECgMPL, which can assess microscopic images of cells and tissue to identify endometrial cancer – one of the most common forms of reproductive tumors – with an impressive 99.26% accuracy. And the researchers say it can be adapted to identify a broad range of disease, including colorectal and oral cancer.

“The proposed ECgMLP model outperforms existing methods by achieving 99.26 percent accuracy, surpassing transfer learning and custom models discussed in the research while being computationally efficient,” said the study's co-author Dr. Asif Karim, from CDU. “Optimized through ablation studies, self-attention mechanisms, and efficient training, ECgMLP generalizes well across multiple histopathology datasets thereby making it a robust and clinically applicable solution for endometrial cancer diagnosis.”

RIP Doctors if AI can do their jobs better
 



RIP Doctors if AI can do their jobs better
I wonder if you have got to ring up for an AI appointment at 8am.
You are 102nd in the queue, your call is important to us...
 
I wonder if you have got to ring up for an AI appointment at 8am.
You are 102nd in the queue, your call is important to us...
Well it would work 24/7 and scale is down to computing resources. How much compute could you buy from one GPs salary?

It's definitely a good use case for these models as a large part of the doctor role is being an encyclopedia, but we'll never fully remove the human element. Best case it speeds up 80% of the process and more people can get help without waiting weeks.
 
ChatGPT still amazes me at how good and simultaneously bad it is for what I do (programming)

It often gives you a useful broad solution to an issue but once you get into the weeds it just hallucinates things. Functions/arguments that don't exist is a classic.
 
ChatGPT still amazes me at how good and simultaneously bad it is for what I do (programming)

It often gives you a useful broad solution to an issue but once you get into the weeds it just hallucinates things. Functions/arguments that don't exist is a classic.

Finding something similar. It gets you up and running and skeleton bits but probe it and it goes off on a tangent and needs reeling in.
A mate seems to find Claude Sonnet more helpful but not given it a chance yet.
 
ChatGPT still amazes me at how good and simultaneously bad it is for what I do (programming)

It often gives you a useful broad solution to an issue but once you get into the weeds it just hallucinates things. Functions/arguments that don't exist is a classic.
Yeah I use it for guidance and tweaking of an already working solution. It helps with the broad work and the fine tuning, but have yet for it to give a perfect solution. But that keeps me in the job so happy days.
 
@fez which model are you using?

Various models honestly. They all fall down in different areas.

Yeah I use it for guidance and tweaking of an already working solution. It helps with the broad work and the fine tuning, but have yet for it to give a perfect solution. But that keeps me in the job so happy days.

I find that for very discrete bits of code, especially things like bash scripts or in languages that don't change much it works far better. Ask it to get involved with anything library related or with APIs that aren't super stable/stale and it just makes crap up.

So broad strokes and very fine details in a limited space its great but I can see a lot of products being build by non-programmers running into huge issues after not very long due to AI going a bit mad and make a spaghetti mess.
 
Chat gpt is doing great things helping me with my university course.

Whilst I don't rely on it to generate a whole assignment for me, it's very powerful in giving me ideas, testing a structure, guiding me to data sources and giving feedback on a draft piece of work.

I use it like a sounding board. I type in the question and see what it produces, then I probe it for more detail in the direction I want my assignment to go, feeding off it incrementally to improve my work.

I can see why universities are worried about it.
 
Chat gpt is doing great things helping me with my university course.

Whilst I don't rely on it to generate a whole assignment for me, it's very powerful in giving me ideas, testing a structure, guiding me to data sources and giving feedback on a draft piece of work.

I use it like a sounding board. I type in the question and see what it produces, then I probe it for more detail in the direction I want my assignment to go, feeding off it incrementally to improve my work.

I can see why universities are worried about it.

That's what I use it for - give me ideas, or help me break the back of some research / help me get started. However the rest of it I use down to my own research, knowledge and seek industry experts. I don't want to rely on these technologies to do the work for me, as ultimately it's my head on the block when I submit that work.
 
That's what I use it for - give me ideas, or help me break the back of some research / help me get started. However the rest of it I use down to my own research, knowledge and seek industry experts. I don't want to rely on these technologies to do the work for me, as ultimately it's my head on the block when I submit that work.

Yeah I agree, it takes away a lot of the mundane initial research phase of a piece of work as it does the job of trawling through the world's knowledge and summarizing it for you. Then you can select the pieces you want and build on them yourself, going back to chatgpt to validate and expand.

It's also quite good at maths. Have to be careful with it but it can do basic statistical work and hypotheses testing quite easily. Again I would ensure I understood the steps it's taking but it really gets you 50% of the way in no time at all leaving more time to polish the work.

I am careful to ensure I reword the work into my own words and don't verbatim copy the structure or order of things that it comes out with. I also search for other sources as it's not very good at finding academic sources so have to be careful there.
 
ChatGPT still amazes me at how good and simultaneously bad it is for what I do (programming)

It often gives you a useful broad solution to an issue but once you get into the weeds it just hallucinates things. Functions/arguments that don't exist is a classic.
it can often be good at debugging or providing some hints which with some more searching can get you to an answer.

It certainly perfect for small, contained code that is somewhat algorithmic, e
find longest common substring or something. Also writing comments znd a dox string with a little tweaking, or some basic UTs

Co-pilot and the like completely fail at understanding complex code bases with lits of interdependencies .
 
For example, I just tried this now with a real use case on o3 mini high:

Me: I need to put a lightning activity feed onto an Aura component in Salesforce, what base component do I need?

GPT: You should use the lightning:activityTimeline base component. This component is designed to display a record’s activity feed (tasks and events) in Lightning, and it can be embedded within your Aura component.

Me: Hmm.... I can't find that - does it exist?

GPT: Salesforce doesn’t provide a public Aura base component called lightning:activityTimeline. The standard Activity Timeline you see on record pages is built into the Lightning Experience and isn’t exposed as a standalone component for use in custom Aura components.

If you need to display activities (tasks and events) in your Aura component, you have a couple of options:

1. Build a custom component: Use the Lightning Data Service (or Apex) to query Tasks and Events and then render them in your own timeline.

2. Embed a Lightning Web Component (LWC): Salesforce has introduced the lightning-record-activity-feed component as an LWC. You can wrap this LWC inside your Aura component (using the LWC-to-Aura integration) if your org supports it and if it meets your requirements.

In summary, there isn’t a ready-to-drop Aura base component for the Activity Timeline—you’ll need to create your own solution or leverage an LWC if possible.


Me: Pretty sizeable error. What made you invent a base component that doesn't exist?

GPT: I'm sorry for the confusion. I didn’t intend to suggest that such a component exists. My initial answer was an error based on a mix-up of component names and features. In reality, there is no public Aura base component called lightning:activityTimeline. The correct approach for displaying an activity feed in Lightning would be to either build your own custom component (using Apex or Lightning Data Service) or to t, which can be used inside an Aura component. Again, apologies for the mistake.


And guess what :

leverage the lightning-record-activity-feed Lightning Web Component

This also doesn't exist.
 
I cannot find reference to that LWC you mention, does that exist?

I apologize for the confusion. There isn’t a publicly documented Lightning Web Component named lightning-record-activity-feed available for custom development. The standard activity timeline you see on record pages isn’t exposed as a standalone base component—either as Aura or LWC—for embedding in custom components.

To display an activity feed in your Aura component, you would need to build a custom solution. This typically involves:

1. Querying Data: Use Apex or Lightning Data Service to retrieve Tasks and Events associated with the record.

2. Displaying the Data: Render the queried data in a timeline format within your component.

If you need help with designing or building such a custom activity feed, feel free to ask!

AI still has some way to go. I got bored at this point and yes, again, 03 mini high.
 
All of them, from every provider. ChatGPT, Grok, Claude - they will all (all models) hallucinate things which simply do not exist.

Not seen it myself on pure coding tasks with o3-mini-high.

I’ve gone through a prompt engineering course though and know how to prime it effectively.
 
ChatGPT still amazes me at how good and simultaneously bad it is for what I do (programming)

It often gives you a useful broad solution to an issue but once you get into the weeds it just hallucinates things. Functions/arguments that don't exist is a classic.

I asked my department (government) on a plugin I needed, things are so locked down I couldn't get it. I basically had an idea I could make it myself, but it would have taken weeks to make.
I gave ChatGPT the challenge and it chucked out a 75% working version, I kept asking it why it wasn't working and with my own knowledge and it's answers I managed to make exactly what I wanted in about 3 days.

Definitely not telling them AI did it. I'll just take the credit.
 
Last edited:
Not seen it myself on pure coding tasks with o3-mini-high.

I’ve gone through a prompt engineering course though and know how to prime it effectively.

Would you mind linking (if allowed) to the specific course?

I'm finding CGPT more and more useful in my work (iOS Developer), too. I use it for various things, from just writing documentation for me, based on the class - to refining my solutions more.

It does hallucinate every so often; but my experiance (over a decade) helps me there lol.
 
Back
Top Bottom