• Modern Chaos
  • Posts
  • MC.9: Emerging Warfare Tech, ChatGPT UI Enhancements, and the Enigmatic $235 Million Bitcoin USB ๐Ÿš

MC.9: Emerging Warfare Tech, ChatGPT UI Enhancements, and the Enigmatic $235 Million Bitcoin USB ๐Ÿš

This week, witness the dramatic shift in warfare with a $400 drone challenging a $2M tank, explore the intuitive interface of SlickGPT, and unravel the mystery surrounding a USB drive housing a fortune in Bitcoin. Additionally, discover advancements in no-code LLM fine-tuning and the performance of Zephyr-7B-beta. Plus, the conundrum of prompt injection and a fresh look at AI-powered customer service with Air.ai.

modern chaos issue 9

Modern chaos is a newsletter exploring tech and AI through the journey of a dev agency shifting from services to product design. We share our notes, analysis and experiments.

In this issue:

๐Ÿ“‹ Notes: ChatGPT keyboard shortcuts
๐ŸŒ€ Zephyr-7B-beta: as good as ChatGPT on AlpacaEval
โš”๏ธ The future of warfare: A $400 drone killing a $2M tank
๐Ÿ“ž Air.ai: 100,000 sales and customer service reps at the tap of a button
๐Ÿ”„ SlickGPT: a better UI for ChatGPT
๐Ÿš€ MonsterAPI: No-Code LLMs finetuning
๐Ÿ” They Cracked the Code to a Locked USB Drive Worth $235 Million in Bitcoin. Then It Got Weird
๐Ÿ›ก๏ธ Anti-ChatGPT: Your digital content filter
๐Ÿ’‰ Prompt injection: a problem I didnโ€™t know I had

Work & Experiments

Notes: ChatGPT keyboard shortcuts
It was only last week that I stumbled upon the keyboard shortcuts for ChatGPT. Although many of them seem less than essential, there is one that stands out for its immense utility: "Set Custom Instructions." READ IT

Tech updates & tools

Zephyr-7B-beta: as good as ChatGPT on AlpacaEval
Take the related performance with a grain of salt. More often than not, stellar benchmarks do not necessarily translate to groundbreaking user experiences. I have a hard time believing this smaller 7B model rivals the prowess of GPT-3.5-turbo. DEMO | TWITTER

The future of warfare: A $400 drone killing a $2M tank
Before the war, a teenager might hope to get a โ€œfirst-person viewโ€ drone for a New Year present. Now they are being used as agile weapons that can transform battlefield outcomes. READ IT

Air.ai: 100,000 sales and customer service reps at the tap of a button
Touting themselves as the premiere AI capable of engaging in realistic 10-40 minute phone conversations, the reality seems a bit too glossy. While the promise is alluring, their sales approach rings alarm bells. In the rush to ride the AI wave, we must remain cognizant of limitations. Here's a thought: if the technology were as revolutionary as claimed, it should be self-evident in valueโ€”no need for the steep trial fees of 30k to 100k. REDDIT | AIR

SlickGPT: a better ui for ChatGPT
For those who find ChatGPT's interface cumbersome, there's a new, more intuitive Slack-like alternative worth exploring. Dive into the full list of features in their blog post. SlickGPT | BLOG

MonsterAPI: No-Code LLMs finetuning
Launch a cost-effective fine-tuning experiment in a few simple steps. This service manages common pitfalls like memory issues, overfitting, and early stopping. However, in my view, the real challenge is in curating the dataset for fine-tuning. MonsterAPI

They Cracked the Code to a Locked USB Drive Worth $235 Million in Bitcoin. Then It Got Weird
Stefan Thomas is the unfortunate soul who forgot the password to a USB drive with 7,002 bitcoins. A team of hackers has claimed they can unlock it, yet Stefan remains hesitant. It's quite the enigma. READ IT

Anti-ChatGPT: Your digital content filter
This is a Chrome extension that acts like your personal content concierge, filtering through the noise and biases of the web according to your own rules. Their example shows how to avoid politically charged tweets; a clever use of GPT-4. GITHUB

Prompt injection: a problem I didnโ€™t know I had
An interesting journey from Benjamin Stein who started seeing prompt injection issues everywhere and tried to come up with a framework using GPT itself to sanitize the user prompts before processing. Turns out to be a super interesting technique for detecting malicious prompts. READ IT


Reply

or to participate.