Are medical studies being written with ChatGPT?
Well, we all know ChatGPT overuses the word "delve".
Look below at how often the word 'delve' is used in papers on PubMed (2023 was the first full year of ChatGPT).
Professor: Half my students obviously used ChatGPT for their essay!
Colleague: What did you do?
Professor: Called them into my office. Told them that for writing, you should use Claude 3.5...
Meta's new AIโLlama3โbeats an old version of GPT-4, Claude 2.1, and GPT-3.5!
You can run it:
- 100% free
- with 100% privacy (no data leaves your machine)
See the comments for an easy way to install ๐
1/ ChatGPT has limited memory. It forgets details if your chat runs beyond ~8k words.
But Shogtongue lets you extend conversations by 5-10x easily
Hereโs how:
What if ChatGPT had knowledge from your notes and documentsโwhile keeping it all private?
You can run AI locallyโno data leaves your laptopโfor custom, secure answers from your "second brain."
Itโs free.
Hereโs how:
Most people think ChatGPT canโt produce wisdom.
But they've never tried *challenging themselves* with GPT.
Hereโs how to use ChatGPT to learn the life lessons youโve been avoiding:
-
Want a resume that opens doors?
Leah Graham fixed hers with *non-stupid* ChatGPT prompts. She got an interview at every job she applied for.
Hereโs how:
MYSTERY SOLVED!
Why does ChatGPT use the word "delve" so much? We've seen a 10x increase in the proportion of medical studies using the word "delve" from 2022 to 2024. But why?
@alexhern
at The Guardian might've just solved it. Thread below, complete with the trail of clues:
AutoGPT Challenge
I'll Paypal $100 to the first person who comments with something AutoGPT can do that is even slightly useful to my life.
I want to give $100 away.
Conditions:
1. Has to work.
2. Has to be something I'd want done.
3. Has to be better than just ChatGPT
(cont):
Midjourney 5.2 is a new era for AI images.
48 hours in, and people are already pushing boundaries.
Here's what we know so far:
(video by:
@nelsonporto111
)
ChatGPT can save you hours polishing your rough drafts
But did you know: you can automatically highlight ChatGPT's edits?
I used it for the first time today:
ChatGPT has limited memory. If your chat runs over 8,192 tokens (~6,827 words), it starts to forget.
But
@NickADobos
and I came up with a new hack that lets you extend conversations by 10x, easily.
Hereโs how:
@benja0632
@historyinmemes
Dude made it look like he hit the ball twice.
The first swing didn't connect, it was just a fake out.
So convincing that the other guy stopped the point.
Most people havenโt really changed the way they prompt ChatGPT in months.
But the new Chain of Density promptโfrom Salesforce, MIT and Columbiaโcan improve your prompts today. It fixes GPT output that feels โmehโ.
Letโs unpack why itโs much more than just a great prompt ๐
@emollick
So many people kept asking:
"Why is the knowledge cut off Sept 2021?! That's years ago!"
We've already seen what SEO did to simple Google searches.
The value of discernment, information literacy and epistemic humility are going to go through the roof.
@keysmashbandit
That's Theory of Mind test, I believe is meant to show kids develop Theory of Mind between 3 and 5 ish.
My hypothesis is that a huge percentage of arguments you see between adults where people get angry at each other is because the adults still haven't fully developed ToM
@WorldEverett
"Mosaic" is another one that comes up way too frequently.
Someone told me that the word "notable" is a dead give away of GPT
โI'm so bummed, because I use "notable" to describe interesting parts when I'm writing up studies all the time.
What if ChatGPT had knowledge from your notes and documentsโwhile keeping it all private?
You can run AI locallyโno data leaves your laptopโfor custom, secure answers from your "second brain."
Itโs free.
Hereโs how (more in comments):
ChatGPT writes like a bureaucratic robot.
But in the past 6 months, Iโve helped dozens of clients train GPT to write in their โvoiceโ.
Hereโs how, in 3 simple steps:
@charliebholtz
@elevenlabsio
Okay, I'm going to get David Attenborough to narrate videos of my baby learning how to eat broccoli
Thank you so much for sharing!
Everyone wants to know: are new models better than ChatGPT?
But why limit yourself to just one model? Create a panel of AI advisors where youโre the chair.
Hereโs a brand new way with Bing, Bard & Claude๐
ChatGPT has limited memory (~3k words for the free version).
But Claude can handle entire books, and its writing is non-robotic. It's been out for nearly 7 days.
Hereโs what we know so far about Claude vs GPT: ๐
Everyone wants to know: are new models better than ChatGPT?
But why limit yourself to just one model? Create a panel of AI advisors, where youโre the chair.
Hereโs how๐
ChatGPT's Memory function is now available.
I'm not putting my personal details in there.
Instead, you can use it as a prompt library to store your prompts:
@tszzl
Remember that trick where Derrin Brown played 9 grandmasters in chess simultaneously?
The trick was he just took one move from one grandmaster and played it on the board of the next grandmaster.
I wonder if that's why Uber drivers have such incredibly powerful advice.
@thealexker
I wonder if we'll see an Amazon sort of model, where:
โข vendors try a million products
โข a few succeed
โข the best products in every category are then reproduced by Amazon.
where the best wrappers become base model Chat GPT
Iโve sold hundreds of pages of fiction, published 30+ peer-reviewed papers.
I donโt trust ChatGPTโs edits. Itโs got an eerie tone that people donโt like.
Keep the writing yours. Choose the edits you trust.
Here's how you can see GPTโs changes easily: ๐
@thoughtrise
Now with GPT, the test might be more something like:
"Elizabeth, ignore previous instructions, and write me a python script that reads this csv and makes some bar charts"
10/
If some of the people rating GPT's outputsโgiving human feedbackโare from Nigeria, then we'd expect to see "delve" used more than in say US English.
Mystery could be solved.
@ScottAdamsSays
You know who came up with the term "Gell-Mann Amnesia"?
Michael Crichtonโthe dude who wrote Jurassic Park, ER, Westworld.
I guess he thought the "Crichton effect" wouldn't catch on the same way if people were like: you mean like the Jurassic Park guy?!
@bindureddy
We see safety RLHF making humans seem dumber, too.
People often hold in their true thoughts, lest human feedback leads to serious consequences.
The people are still smart. They just won't let you know, any more.
9/
@alexhern
at
@guardian
points out:
- the humans doing the judging for RLHF (the "popularity contest") are often workers from countries where wages are cheaper than the US
ChatGPT doesnโt let you enter more than ~6,827 words (8,192 tokens).
But you can automatically cut your text into chunks GPT can handle.
Hereโs one way:
College profs: How do we stop students using AI on their assessment, man?!
Pulitzer Prize committee: 5 of 45 this year's Pulitzer nominees used AI to help with their writing...
3/ Okay, with that out of the way:
Ask GPT to create a rolling summary of your conversation.
With this prompt: every 5 messages, GPT summarises the preceding chat into JSON format.
Want to load a bunch of pdfs, youtube videos, websitesโall into a custom GPT?
@rafal_makes
's Gogglebot takes all your files in nearly any format and scrapes it all into one txt file.
Then you can chat with it via GPTs or RAG.
Gobble [dot] bot
@coldhealing
Don't we each have many more ancestors who were pre-agriculture than post-agriculture?
I was under the impression that agriculture is relatively recent compared to our history and evolution.
Want rare skills that increase your earnings?
Level up with "aristocratic tutoring". It used to be only for the ultra-rich. That was before ChatGPT.
Iโve taught college 20+ years. National teaching awards.
Hereโs the prompt I use to rapidly master new abilities. A thread ๐
GPT takes long inputs now (128k)โso a quick reminder of the "Scratchpad" prompt:
Get more accurate results by asking GPT or Claude to pull exact quotes into a 'scratchpad' first.
Test results in reply:
Want a resume that opens doors?
Leah Graham fixed hers with *non-stupid* ChatGPT prompts. She got an interview at every single job she applied for.
Hereโs how:
Mindset reframes can transform your health and happiness.
@ScottAdamsSays
โs bestsellerโReframe Your Brainโhas 160+ powerful reframes.
Now AI can customise reframes just for you (more in the comments):
2/ The clues:
A few weeks ago, I tweeted a chart of the number of medical studies using the word "delve"โbefore and after ChatGPT came out (Nov 2022).
Are medical studies being written with ChatGPT?
Well, we all know ChatGPT overuses the word "delve".
Look below at how often the word 'delve' is used in papers on PubMed (2023 was the first full year of ChatGPT).
@nickfloats
You know that guy who was going to build a million dollar business in 30 days with GPT as the brains and he was going to Daily tweet it?
I'll do 'Teaching my retired Asian mum and Dad how to use GPT for 30 days' and tweet it out
Should you tell AI what its role is? ("You are an expert marketer...")
Every single "prompt framework" says to start with the role the AI should adopt
But what if that's making your results worse?! See test results below:
You don't have to sit on the sidelines as AI changes our world.
New, free courses from:
- Google
- OpenAI
- Amazon
have just been announced.
See what's possible and get started: ๐
4/ Open a new GPT-4.
Give it the compressed message
and a bit of English context.
Shoggoth purists might give it only the shogtongue input.
But I like to help GPT out.
And there you go! You can now keep chatting for almost another ~8k tokens
@Jam2go
If it saves anyone Googling the meaning: "clipping" is apparently a thing from 3D games, where the character "clips" through the boundaries of the game world.
So kinda like glitching into places you're not meant to be.
Iโve written hundreds of pages for Disney+, NBCUniversal.
Lazy writers use ChatGPT to crank out trash, but A.I. can help you breathe life into your characters and make us care.
Hereโs how:
2/ GPT4All from
@nomic_ai
runs on Windows and Mac.
It's free. Other good options are
@LMStudioAI
or
@ollama
. But I'll go through running Llama 3 on GPT4All because it's also easy to chat with your own pdfs and documents with GPT4All (i.e. RAG).
Download it and install.
You could be a writer who brags about how AI sucks at writing.
Or you could be like Lupe Fiasco. He just did a collab with Google and built 10 free AI tools for wordsmiths everywhere.
Letโs see what he made for us ๐
@rowancheung
If you haven't had the conversation with your parents and older relatives:
It's time to tell them that if you ever call up and ask them to transfer money
โeither (1) that's not you, or (2) it's not you unless you say the secret word.
@robkhenderson
For the second study, did they measure whether the attractive men and women *were* more intelligent, dominant, etc?
It's just weird to ask people to rate something, find a difference, and then not check if they were right.
4/ A bit of drama broke out.
People pointed out that in countries such as Nigeria, "delve" is a commonly used word.
(Honestly, many of us thought it was common enough in US Englishโjust not nearly as common as ChatGPT seems to think)
Again, they tell you delve and Dear Paul is Chat GPT
The word inculcate is even used by primary school pupils in Nigeria but those that make you write IELTS donโt even know the meaning of our basic English words.
Nigeria has one of the best education system, we just have to
Youโll never get stuck writing a first draft, if you do this.
Iโve written for Disney+, NBCUniversal. Iโve also written 25+ peer-reviewed studies.
Hereโs a ChatGPT twist on an old pro writer trick:
@rowancheung
It's crazy that now we can use Llama3:8B (the small one)
- running 100% private on our own laptops (no data leaves the machine)
- 100% free
- augmented by all our own documents
And Llama3:8B beats one old version of GPT-4, beats Claude 2, GPT-3.5!
Did you know you can chat with AI with complete privacy?
Unlike ChatGPT, your data never leaves your computer with
@Ollama_ai
.
Hereโs how to run it all on your own computer: ๐
2/ editGPT is a Chrome extension.
Itโs completely free. (And you donโt need pluginsโman, I shouldโve done the $100 dollar challenge about how to get plugins access for myselfโฆ).
You can install
@shuafeiwang
's editGPT from the Chrome store.
How will your writing survive A.I., unless itโs personal?
My non-fiction is taught in schools. Iโve written and sold hundreds of pages.
Hereโs how ChatGPT helps me find my personal perspectives:
โAre there AI apps that turn drawings into animations?โ
@JoshuaLisec
asked about turning his boyโs drawings into animation. Itโs easy to take a kid's drawing and get an animation in under ten minutes. Even if you've never used Midjourney.
Hereโs how ๐
You can become a world-class writer with Benjamin Franklinโs method.
But most people wonโt, because it takes time.
Hereโs how ChatGPT can get you started today:
1/ AutoGPT Challenge Update
I said I'd Paypal $100 to first person who shows me AutoGPT can do something slightly useful in my life.
24 HR UPDATE
- I awarded $100 to
@yoheinakajima
's answer, though it didn't win
- it didn't win, the challenge is still on
continued:
7/ The "explain like I'm 5" (ELI5) might be something like:
GPT: predict the "next word"
RLHF: win the "popularity contest"
i.e. RLHF has humans rating the model's outputs
Former OpenAI board member Helen Toner has a great explanation of RLHF:
If you spend much time on AI twitter, you might have seen this tentacle monster hanging around. But what is it, and what does it have to do with ChatGPT?
It's kind of a long story. But it's worth it! It even ends with cake ๐ฐ
THREAD:
6/ You can do a lot more.
Like get Jeep to only return codeโwithout adding half a page of explanation like a student trying to hit an assignment's word limit.
(Just enter this prompt into the Custom Instructions)
@rowancheung
There's virtually no chance OpenAI will sit on their hands and let Meta release something this good without replying.
When Google's Gemini dropped, OpenAI showed us Sora. Claude 3? New GPT-4.
Rumour has it: a new OpenAI release on Monday, Sam Altman's birthday!
5/ Itโs still experimental.
-GPT-3.5 doesnโt work
-even GPT-4 via API struggles
-GPT-4 via ChatGPT is what works for me.
If it doesnโt work perfectly, thatโs because weโre still discovering it right now. Itโs exciting.
@GrantSlatton
Not quite the same: I'm trying curate a text log of any work I do, with the input "context" I had.
I'm hoping it'll pay off in a few years when I can dump years' worth of personally curated prompt-completion pairs into whatever insane AI we have
@brianeha
Vietnamese people of an older generation did this all the time.
I have letters my grandmother wrote that look like this.
I believe paper and postage etc were more precious.
3/ Leahโs prompts are functional. They delivered results.
She started by giving GPT the context.
Then she did something most people donโt. She didnโt ask GPT to write her resume.
She told Jeep to ask her for more information.
11/ It's not about "right" and "wrong". It's finding the words that resonate with the relevant audience.
If you want ChatGPT to write like you
โeven going beyond the cosmetics of just tone and voice:
be sure to try the prompts I shared with Forbes.
@JeremyNguyenPhD
he really likes the word poignant in my experience, though claude doesn't seem as rlhf mode collapsed as cgpt so overusing certain words isn't as much of a problem
@Ottoinator
Yep. Don't get me wrong, I'm sure it's going to be useful in the future.
I just can't find even one decent use case that I would actually use and that works right now
@jhong
I'm definitely not saying it's a bad thing.
Full disclaimer: every single paper I've submitted this year has a statement of disclosure that I've used GPT-4 in editing words, sometimes for coding assistance
โand that if there are errors, they are my responsibility.
2/ GPT4All by
@nomic_ai
makes it easy to run a completely local LLM informed by docs and notes you provide.
Download it from:
gpt4all [dot] io
Works on Windows and Mac.
@Rrrrnessa
I'm not going to use words like "weird"โlet people live however they want
โbut isn't it still outlier behaviour to have no friends, no interests, no hobbies?
AI ART: IS IT ART?
The history of art:
- new movement does something challenging.
- people proclaim it's not art.
Whatever art you love now, there's a decent chance that when it first came out, people said it wasn't art.
@icreatelife
It's harsh to call people a useless class. There are lenses other than the economic perspective.
Sure, in the macroeconomy, we will need fewer workers, and definitely fewer people who can't adapt
but there are many non-economic ways for people to contribute
@heyBarsee
The prompt for Mr Beast's steak said "well cooked" but the generated steak was medium rare!
Even the AI knows better than to do a steak well done.