AI is Better at CAPTCHAs Than You

Welcome to edition #24 of the “No Longer a Nincompoop with Nofil” newsletter.

Here’s the tea ☕

• Women walks again thanks to AI 🚶‍♀️ 

• GPT-4 is being used for content moderation 💻️ 

• CAPTCHAs have been made obsolete 🤖 

If you’d like to explore how AI can enhance your business, reply to this email or contact me [email protected]. For extra reading, scroll to the bottom.

Enjoy 🍷 

Identifying humans just got harder

It’s finally happened. CAPTCHAs have become redundant. A new study shows that ML-powered bots are better at solving CAPTCHAs than humans. Not only better but also faster. CAPTCHAs are one of those things that weren’t exactly a great way to determine if someone was human, but since they worked for so long we used them. There was no alternative.

If you’ll remember back to the release of GPT-4 way back in March, which now feels like a century ago, OpenAI touted releasing a multi-modal model soon. We are still yet to see any evidence of this model besides the initial demo all those months ago. Why? Someone got access to the GPT-4 multi-modal model and it easily solved CAPTCHAs. Some speculate that OpenAI still hasn’t released it because we simply have no alternative to figuring out if someone is human. Can you imagine if OpenAI released this model and people started using it to scam people? I wouldn’t be surprised if this is a contributing factor to them not releasing it.

If you haven’t watched the news recently, Sam Altman, CEO of OpenAI has also been talking about Worldcoin. What’s Worldcoin? Simply put, you scan your eyeball to verify you’re a human and you get some crypto in return. This is supposed to connect to a digital ID which you can use to verify your humanity online.

Somehow, I think this won’t work. I’m not sure people are ready to scan their eyeballs like they’re in Blade Runner to verify who they are. I know I wouldn’t. Companies have enough of our data, scanning my eyes would feel beyond creepy, but that’s just me. Each to their own I suppose.

GPT-4 will wipe out moderators

OpenAI announced they’ve been using GPT-4 for policy development and content moderation. This allows them to update policies and label content much faster. It can’t be that much faster right? OpenAI claims it reduces moderation time from months to hours. They also claim it creates more consistency, streamlines a faster feedback loop and “reduces mental burden”.

I think this may be in reference to this article, highlighting that the people that helped moderate content for OpenAI became depressed and anxiety-ridden from reading all of the insane and vile things models would output. If this problem is as big as it seems, then this will be very helpful.

My only concern, like anyone else I believe, is letting an AI simply do all this moderation for us. Are we ready to remove the human equation from such an important process in so many different industries and businesses? I’m not sure. But according to the data, GPT-4 is already just as good, if not better. Soon enough, it’ll far exceed human performance. The inevitability of AI.

AI x Health is the future

A 65 year old stroke patient in the UK was one of the first people in the UK to trial the NeuroSkin “smart garment”. The trousers help stimulate her weakened leg allowing her to walk in a natural motion. Stroke patients can suffer from some form of paralysis which makes life after suffering a stroke very hard. The Neuroskin is worn as trousers and a vest and constantly “listens” to the impulses being sent around the body. The system tracks the impulses from her healthy leg to the brain. It then translates the same impulses for the weakened leg, essentially mirroring the process and guiding the brain to simulate the leg appropriately.

This will completely change how we manage stroke patients and the quality of life they can have after suffering a stroke. AI used in health, stimulating the brain to perform certain tasks and actions is so very powerful and this is just a glimpse into how it will make life better for a lot of people. The intersection of AI and health is definitely one of the most exciting industries moving forward. We are still so, so early.

Extra Reading

• Platypus is a new LLM topping leaderboards [Link]

• Google released a new dataset to help see advances in document understanding and parsing [Link]

• MultiOn API released this week [Link]

• This paper discusses using Code Interpreter to boost mathematical reasoning. Link to paper [Link] Link to thred breakdown [Link]

• AI powered cleaning robot is coming [Link]

• Dreambooth dataset is now publicly available on HF [Link]

• Text2Material is a genAI tool that allows 3d artists to generate PBR materials from text descriptions or reference photos [Link]

• Whatsapp is experimenting with text-to-image generations for stickers [Link]

• A federal database of AI use cases for the department of health and human services [Link]

• DeciCoder 1b - a local LLM that writes code for you [LInk]

• OpenAI released a video with their CTO Mira Murati on how ChatGPT works [Link]

• The allen insititute is developing their own LLM called OLMo and they released the dataset for it called Dolma. It’s 3+ trillion tokens [Link]

• Structured LLM APIs with Thiggle [Link]

• Someone is creating a conversational music tutor, looks cool [Link]

• Runway released a streaming service?? Find people building with AI on runway. Available in browser and iOS

As always, Thanks for reading ❤️

Written by a human named Nofil