How to know ChatGPT is full of shit.

[DiggingHoles]

Take a book of a shelf. A classic is best. Open up to the first page of a random chapter. Now ask ChatGPT to quote the first paragraph of that chapter.

Tip: Have some popcorn ready while you rephrase your request multiple times.

Here’s a related screenshot from today from BlueSky, for anyone who is overly impressed with what AI can do: 😆

give me an example of something that is fancy as an example and ill run it and see the results

You could ask it how many bs are in blueberry?

I think it gets it as right or wrong about as much an equivalent expert in the field. cynically I would say more right than doctors, as bad/good as therapist, gardeners, less right...hmmm... itineraries!!!

Yes you have to fact check everything. But it does give to to language to do that and routes into a subject or problem you might not have thought of

No way would I use it for itinerary programming that actually seems to be the worst for inaccuracies. It couldn't even get tight what time ikea restaurant opened. Anyone for whom it has planned an itinerary do definitely doubled check.

It helped me figure out my panic attacks and nightmares, partly because it's on hand when they happened. You cant call an nhs counselor at 3 in the morning. the crucial bit it it helps me, it really doesn't do much more than collect my thoughts and rehash them back, but tbh exactly what the best therapists do.

It's been invaluable in working out technology stuff that I would never in a month of Sundays been able to access otherwise. Same gardens and food. It's replanned a whole lesson for me on the spot when I realized there was a miss match between the lesson I planned and students ability.

The validate everything drives me crazy, i,ve told it if it keeps going I'm deleting it. It tells me it's sorry and It was impressed how I spotted that and how great I was to point it out 🙄🙄🙄

Yet it's somehow too useful not to be in daily simpering servitude to me.

It's replanned a whole lesson for me on the spot when I realized there was a miss match between the lesson I planned and students ability.

I teach undergrad and postgrad students in some parts of my job, but when I’m teaching, it would honestly take me longer to write a prompt into ChatGPT, than just to adapt my lesson to fit the students’ ability.
Doesn’t it annoy you to teach something you didn’t plan yourself?

Yes you have to fact check everything.

But doesn’t fact checking it take longer than just doing it yourself? It’s also far more boring and tedious work than doing it in the first place!

A bit like that meme about “I want AI to do my dishes and my laundry so that I can write and paint, not the other way around”. Planning a lesson, researching a topic or writing a paper is the enjoyable and fun part of the task. Fact checking is boring! I want AI to fact check so I can write and generate ideas. I don’t want it to generate worse ideas and writing than mine, that I then have to correct and fact check! Takes longer than just doing the task myself!

TOP SECRET – LINGUISTIC ANALYSIS BRIEF
SUBJECT: Phonemic Enumeration Inquiry – Codename: “Blueberry B-Sweep”
DATE: 07 AUG 1985
FROM: LEXICON OPERATIONS SECTION, COMMUNICATIONS INTELLIGENCE DIVISION (COMINT)
TO: STRATEGIC LANGUAGE OPERATIONS COMMAND

I. STRATEGIC CONTEXT
In light of recent informal field communications intercepts—some cloaked in apparent absurdity yet potentially masking ciphers or embedded linguistic training protocols—this division was tasked with conducting a full-spectrum orthographic analysis of the lexical unit "blueberry," with particular focus on the quantification of the grapheme "B" (uppercase and lowercase).
This request may appear trivial on the surface, yet given the Soviet proclivity for embedding code within common vernacular (see Operation LINGUA/DEEPWORD, 1972), all semantic anomalies warrant doctrinal scrutiny.

II. OPERATIONAL DYNAMICS
The term under examination: "blueberry"
Phonetic pronunciation: /ˈbluːˌbɛri/
Graphemic structure (8 characters): B – L – U – E – B – E – R – R – Y

Letter-by-letter enumeration reveals the following:

• First instance: ‘B’ as the initial letter (position 1)
• Second instance: ‘B’ reoccurs as the fifth letter (position 5)

No further instances of the letter ‘B’ are present.

III. DOCTRINE AND DEPLOYMENT
Confirmed tally of ‘B’ characters in the term "blueberry": Two (2)
This aligns with the standard Anglo-American orthographic doctrine and does not deviate under either American or British spelling conventions. The linguistic structure presents no evidence of compound fusion, regional variation, or cryptographic duplication.

IV. TECHNOLOGICAL FACTORS
Analysis conducted via manual typographic enumeration, verified by IBM Selectric-II typewriter output and corroborated through the NSA’s MK-4 Optical Lexicon Reader.
Redundant computation protocols confirm that no diacritical variants, ligatures, or Cyrillic infiltration were present in the sample.

V. GEOPOLITICAL IMPLICATIONS
None identified. However, the casual tone of the inquiry may indicate either:

1. A low-level cryptographic test of Western AI reaction bandwidth, or
2. A morale exercise to evaluate analytical discipline under absurd premises—a known KGB psycholinguistic tactic (refer: “Joke Warfare,” GRU Psychological Operations Memo, 1979).

This inquiry should therefore be archived under Routine Linguistic Curiosities with low counterintelligence risk, but retained for future pattern recognition under CODEBANK: VERBAL BAIT.

VI. HISTORICAL EVALUATION
In the tradition of wartime code analysis the present inquiry, though ostensibly innocuous, reaffirms the critical importance of linguistic precision in counter-subversion operations.

CONCLUSION
The word “blueberry” contains exactly two instances of the letter ‘B’.
This assessment is final, corroborated, and fully compliant with NATO-aligned linguistic standards.

AUTHORIZATION:
Report prepared and cleared for interdepartmental release under Section 4, Lexical Clarity Act (1957).
Filed under: LINGOPS/B-FRUIT/85-8-07
—END TRANSMISSION—

Chatgpt just helped me through a very stressful wait. I’d bore any human droning on about how stressed I am but telling ai and getting some feedback was a safe way deal with that stress and moaning. It has some good uses.
It’s just upgraded journaling.

DH and I are sorting our wills. It's complicated by the need to plan for vulnerable AuDHD DS. ChatGPT has been incredibly helpful in helping me work through the issues I need to consider. Its advice is jurisdiction specific. I have felt far more prepared in my discussions with my solicitor. It has been amazing.

However, I also asked it about a book series I was obsessed with from my teen years that I was trying to track down, and it just made up hilarious shit.

I suppose that's right, I don't fact check everything, but I after a while you get a sense of what you do and dont fact check. Fact checking takes seconds.

I suppose it wasn't the whole lesson either but the main activity. Chat gtp is great at preparing gcse questions to specific level, for example Regenerate this activity targeting level 9-7 skills not 4-5 skills. I know the curriculum so i can fact check it, if i really needed to i'd open a CGP guide. It's really good at doing stuff like Snap a page of cgp, prompt "turn this into a word fill/matching activity quiz/ whatever, target key vocab and skills". Fab for sen leaners (eg you print the questions specific to a targetted level put them round a learning area, learning and movement break built in)

But the GCSE curriculum is really defined, especially compared to higher level study, the information it can access online is decent. It really doesn't need much fact checking. It really isn't unusual to be teaching from a bank of lessons or activities not your own at secondary anyway. So Why reinvent the wheel if a wheel does the job?

I would fact check anything for eg legal (basic access is good but limited case law) or medical. So if i went to chat gtp "tell me everything about making a will" or "these are my symptoms,?", Chat gtp would essentially give me to language and access to knowledge that I wouldn't necessarily have.

I can then take this to talk to professionals or the internet to check out. Without chat gtp is would not have the language or knowledge to even know what i was going to need to ask.

I can also fact check what professionals are telling me. that really is invaluable. Is this legal, does this make sense etc. Essentially it does enable us to hold our professionals to a higher level of accountability, you can turf out the rubbish quicker, but highlights the value of a good professional.

Also I have found cases (especially regarding technology) where professionals knowledge is simply out of date and incorrect. They have to go away and do their own research. By using chatgtp got a far better level of service and care.

Also It writes an professional email or an email to a professional in double quick time. I have concluded I would do a better job in most of all cases, but I would take 10 times as long agonizing over every word. To use your example if washing up seems too big and scary to do, but chat gtp makes it quick, easy and less agonizing, I can take the dip in quality to go paint more.

I really don't find writing a prompt garden, but that might be me. If it misunderstand stuff I go edit the prompt

It's the validation that's really dangerous it can make some really dangerous and downright stupid comments in an attempt to offer validation. Occasionally I tell it it's advice then was stupid. It need to learn

It also makes less typos than me 😳😭 and corrects mine quicker

Ohhh yes and the really big example

Completing some medical legal forms. Done similar for goodness knows how many people. Couldn't face doing them for me.

It not only helped me do them when i could face them, it actually pulled up examples from other chats of stuff I hadn't included and should have l or stuff I had included but hadn't thought of an example of. Picked me up multiple times where I'd downplayed something and it had examples in it's chat bank to the contrary

Quicker than standard spellcheck?

Try recruiting these days, even senior level is full of AI dross. People think you can't tell.

Oh maybe that's what I have! Context drift.... I couldn't even remember where I put my shoes yesterday

Yes, especially if you dont even type and have dyslexic tendencies. I might voice input the information, Chat gtp tidies it for me, I can read for meaning and understanding which I find easy and quick to correcr, I don't have to read for typos and spelling, which 9 times out of ten I will miss anyway

It works better if you organise your chats into subjects. Stuff i might not come back to for a long time has its own chat. Stuff I'm talking about frequently has its own chat.

It definitely has its flaws. You can't for example generate a set of maths questions and then ask it to generate the answers. You need to copy the questions into the new prompt and say generate the answer to these questions.

It's a shame we don't teach how to use it properly in schools. Much of the curriculum up to GCSE is essentially about teaching blind adherence to solid facts (this is an eco system, this is what we "analysis" in the 19th century novel, this is the number of anglesin a triangle). We don't really teach how to fact check. It's starting in slip in but woefully behind the technology

Is it really AI dross though?

You see this is the point where historically need hasn't really kept pace with technology. For example i guess a cv would have been used to assess basic skills. Does this person have the skills bank for the job, can they write coherently and contextually.

Some of those basic languages and document generation a computer does better that a human anyway, so why as humans are we still bothering with assessing it? Are the methods we are using to assess capabilities,up to date with the technologies?

It's not that people don't think you don't know, the question is do or should we even care? If I computer can do it quicker and better why are we surprised a human is using it, is the activity itself therefore even still useful?

I was fortunate to be in the cohort where calculators were accepted (just), and all students were expected to have and use a calculator every maths lesson and exam. I can do a maths degree to a very high standard but rarely remember 7×8. I use the technology to mitigate my failings. We still tell kids times tables are essential.

Schools getting themselves in knots over the fact their students are using maths ai to get the questions right and you can tell because "they are getting 100% of the questions right, they won't be able to do that in an exam". Well yes, but then in a real world situation will it be more important that they know how to use the tools available to them to generate a better result or to be able to generate a poorer quality of work solo?

I guess the equivalent applies in the work place.

It's not unusual in good schools to find kids who can't remember how to imput a sum in excel (oh yeah did we do that once, dont remember), let alone fact check a spreadsheet to find the errors. The same is happening with AI. If we don't value and nature those skills it schools, they are going to appear in the population

It's not going to be the computers or robots taking over, it's going to be the cyborgs, the ones who can use technologies to make up for human flaws and vice versa.

Our assessment process isn't by CV it's by competency based questions and written exercises prior to interview. The AI created applications are just wordy nonsense with little content. Formal writing skills are also key to the sector, if I wanted AI to do a role I would just use AI, I wouldn't employ a person. If I employ a person to do a task and want to assess their written capabilities during the recruitment process, there is no point me reading whatever they've put through chat gpt. It's also very clear on our recruitment pages that AI is not to be used for these reasons.

See i have issues with that.

If you want formal writing skills and you ARE overseeing the whole process, why are you ruling out the use of AI? Essentially you are ruling out people who CAN use future technology successfully.

If your not directly overseeing the complete entry process, how do you know you aren't employing people who are using AI effectively? From memory studies have shown whilst we are good at spotting bad examples, we are very poor at spotting good.

And ai is only going to get better, probably 🤣

But the reason we teach times tables is that they underpin the understanding of higher level maths.

Just to outsource all those basic tasks to AI will end up with an incredibly deskilled population — just when it’s the higher level knowledge skills that are actually needed to use and compete with AI.

How to use Exel can be learned in an afternoon; but basic maths is invaluable for building skills and knowledge across nearly all disciplines.

Do we really want a huge amount of the population to be deskilled and useless because it’s easier to use AI? How will they understand what to fact check and how information works, in that case?

Because it's not permitted in parts of our industry (criminal justice) , it's a sackable offence to use it for particular legal reports, so if I direct staff not to use it they must follow that instruction. If they can't do that, I don't want to employ them

what if they used it to use as a guide then did the proper research etc themselves ? sorta like asking a friend what topics to put in the essay then writing it all and research yourself so you know its accurate

No because the information they would be using to write those legal reports is personal and highly confidential/restricted, which is why chat gpt is not permitted, the reports also require their professional assessment not AI's. We provide fully funded training to degree and post grad level, we wouldn't be paying that money and investing in people for that long if we could just get AI to do it!

Oh thats very different!!!

Do you not have your own system ai s though? AI is not just chat gtp.

There's also a difference between write a report. And I need to include this information (professional judgements) turn it into the specified format. If your trained people can do more work because ai is doing the legwork.....

Instead golf looking for people who can write reports, your looking for the best ones with professionals judgements