Today, we’re not going to get ChatGPT to impersonate someone famous for the newsletter intro. It’s straight to business.

Time: AI o’clock.

News Headlines

• New Open Source AI Model Alert 🚨

• Getting the Impression ChatGPT Is Getting Dumber? You Are Not Alone

• It May Be Stupid Sometimes, But It’s Still Taking Jobs 🥲

• AI in Healthcare Segment 🩺

• AI’s Cool And All But Apple Has Other Priorities 🙅‍♂️

Bodacious AI Stuff— 1 Prank, 1 Practical Edition

• Gamma - Create amazing presentations with great design in a fraction of the time. Looks like tome.app might have some comp 😳

•  ThePaige - I think everyone should be able to make a website, coding skills or not. So do these guys, you just gotta pay first. But they’re AI-generated copy looks pretty good to me.

• ROAST - AI optimizes your dating profile. I definitely don’t need this at all despite what people may say

✈️ An AI Flyby for the Week ✈️

New Open Source AI Model Alert 🚨

There’s a lot of good open source models out there that are closing in on OpenAI. Here’s just a couple so far (Developer, model name):

• MosaicML - mpt-7B

• Meta - llama

• Stanford - Alpaca

And now another one joins the ranks. The UAE recently released Falcon, a fully open language model that’s on top of the HuggingFace (open source AI model hosting platform) leaderboards.

Falcon is both faster and more efficient that GPT-3, costing ¾ of the training compute budget while running at a 5x speedup. It has 40 billion parameters and was trained on 1 trillion tokens using 384 GPUs on AWS over two months.

Basically, it’s top of the leaderboard for a reason. Falcon is good. Really good.

Let’s see what kind of surprises the open source community has in store for us as they start developing and improving Falcon.

You might be wondering, why are open source model releases even a big deal though? It’s because there hasn’t been an open source model that outscores OpenAI across the board, but once there is, that would upend the current power structure as hundreds of thousands of developers flock to the new, best model.

READ MORE HERE

Getting the Impression ChatGPT Is Getting Dumber? You Are Not Alone

Many people are taking to social media, claiming that ChatGPT is dumber than it used to be. People feel like it doesn’t follow instructions as well, it’s generations are generally less accurate (especially for code), and cannot retain information as well.

People are starting to blame OpenAI’s latest batch of updates for the new problems cropping up. Apparently, someone was so convinced the latest batch of updates was to blame, that they ran an experiment between two different versions (perhaps using the developer API?) and concluded the following about the updated version:

• it generated more buggy code.

• it was also worse at reviewing code.

• it hallucinated way more when confused.

Another theory about the recent observed dip in performance is that OpenIA’s new safety measures and content moderation / usage policies are deteriorating ChatGPT’s utility in unforeseen ways.

OpenAI is trying to tackle the problem of ChatGPT’s little hallucinatory episodes with something called process supervision (where you essentially reward completion of smaller subproblems instead of just the final outcome). It remains to be seen whether this methodology will help fix ChatGPTs hallucinations. I sure hope so because I hate when it lies and I don’t even notice till later.

But while people are postulating about what could be causing this perceived deterioration, I got to the bottom of the matter through my innovative thinking and stunning use of intellect. Behold

Well, that settles it.

Bonus Section

If ChatGPT really is losing its edge, perhaps it’s worthwhile learning how to use it better to compensate. OpenAI recently released a new guide on how to effectively use ChatGPT to the its fullest. The main takeaways are:

1) Break up tasks into sub tasks (specify steps). For example if you were using it to write a paper, go section by section rather than trying to get it to spit the whole thing out.

2) Try and be as specific as possible

3) ChatGPT likes EXAMPLES. Don’t overflow the context window, but don’t be sparing either if you have a lot of them.

READ MORE HERE

It May Be Stupid Sometimes, But It’s Still Taking Jobs 🥲 

According to data from Challenger, AI was responsible for 4,000 job losses last month.

In May, employers in the United States reported over 80,000 layoffs, marking a 20% increase compared to the previous month and a 287% increase compared to the same period last year. 5% of these 80,000 cuts are being attributed to AI.

The job losses due to AI are likely only to grow if you consider that we are still in the early stages of AI adoption. Although, AI seems to be all we hear about nowadays, it was found that only about 14% of US adults surveyed have actually used ChatGPT and around 48% actually know what it is. Judging by these numbers, AI market penetration and disruption still has a long way to go.

But remember, AI is not just here to take jobs, it’s here to create millions of them as well. Innovative tech has a history of destabilizing the job markets, but new skills come into demand and new opportunities for work open up. That’s how it’s always been and I don’t see a compelling reason for that paradigm to up and disappear just yet.

READ MORE HERE

AI in Healthcare Segment 🩺 

Let’s run through two AI achievements in the realm of Health Care and the Life Sciences this week.

1) Carbon Health Technologies, a network of renowned clinics, introduced a new tool that harnesses text generative AI models to revolutionize the creation of medical records. The intent behind this technology is to reduce the administrative burdens for doctors so they can primarily focus on administering the best patient care they can rather than worrying about paperwork.

Here's the rundown on how it helps:

1. Patient consultations are recorded and transcribed with high accuracy

2. The transcribed text is augmented with supplemental data, such as lab results and physician notes, culminating in a comprehensive overview of the patient’s visit.

3. Utilizing GPT-4, the comprehensive summary is then used to develop precise instructions for patient care and generate appropriate codes for diagnoses and billing.

And get this — A whopping 90% of the submitted transcripts demand no modification from the healthcare provider. Pretty rad if you ask me.

READ MORE HERE

2) Using an AI algorithm, researchers obtained 12% improved accuracy in using genes to predict the development of certain health problems. Imagine what a close to 100% accurate prediction algorithm could do for patient treatment!

READ MORE HERE

AI’s Cool And All But Apple Got Other Priorities 🙅‍♂️

At their recent developer conference, Apple shocked onlookers and attendees as they proceeded to unveil not a single major AI update (except for maybe improved AI spell check). What’s the deal with that?

It looks like Apple might be sticking to improving their hardware for the time being as they bet big on the the future of the Metaverse. I’m under that impressions because of the grand unveiling of their new AR headset, “Vision Pro,” for a whopping $3,499. I could buy two decent MacBooks with that kind of money.

And that’s not all Apple had on the docket. They also revealed:

• M2 Ultra, Apple’s new super chip

• NameDrop — use the power of Airdrop to exchange contact info through a phone tap (finallleee)

• Journal, an ML-powered personalized journalling app (interesting, not really the journaling type though)

READ MORE HERE

Parting is Such Sweet Sorrow (But Please Keep Reading LOL)

Well that's a wrap on the newsletter. Before I launch into my ending monologue, please take a moment to drag this email to your primary, so that you never miss it each week! Thank you!

Let me know how I did by filling out this feedback form or just reply to the email like you would any other. If I did well and you're in the giving spirit please support the newsletter by buying me a coffee or donating to my gumroad page. Most of the money you donate to me will go back into purchasing AI products to try out and review on the newsletter. Bernie says it better:

Also if you have a product that you would like to be featured on the newsletter remember to add it here!

Lastly, if you'd like to refer a friend just tell them to visit chatgptnewsletter.com to subscribe! Or you could always just forward them the email without any subject header. I'm sure they they would love that.

Cool Things You Should Try/Buy

(Newsletter 1) ==> Consult domain experts in your browser with ExperAI

(Newsletter 2) ==> buy a talking book from Konjer

(Newsletter 3) ==> Pay for DoNotPay lol. Irony aside, it really is pretty cool.

(Newsletter 4) ==> Multion.AI - not your average burger buying extension.

(Newsletter 5) ==> Why settle for regular email when you could have Intellimail?

(Newsletter 6) ==> The AI app store where you can Cookup just about anything

(Newsletter 7) ==> A cute, fuzzy AI-powered online meeting summarizer: Otter.ai

(Newsletter 8) ==> Tired of taking hours to find the perfect online purchase? Getproduct.help’s got you covered

(Newsletter 9) ==> Turn your Scribbles into art with ScribbleDiffusion

(Newsletter 10) ==> Ever wanted a search engine for your entire online life? Rewind.ai

(Newsletter 11) ==> Your Online Image Studio 🎬: Clipdrop

(Newsletter 12) ==> Use Generative AI to create online courses on any topic you can think of. For example, you can create lesson plans about yourself before you get famous. Learn.xyz

(Newsletter 13) ==> Replace Siri with ChatGPT— HeyGPT!

(Newsletter 14) ==> Autonomous AI Agents in Your Browser— AgentGPT!

(Newsletter 15) ==> Create your own AI music with musicfy

(Newsletter 16) ==> Detect AI written content more reliably with GPTZero’s browser extension.