Headlines You Should Know
OpenAI Allows Customization of GPT-3.5 Turbo, Adds ChatGPT Enterprise
There’s a lot of power behind base AI platforms like ChatGPT. Many businesses are opting to train those general models on their own data, “fine-tuning” them to better fit their needs. The same concept can be applied on a smaller scale by uploading notes to custom instructions so the model remembers certain information or formats responses to match a specific tone or voice. OpenAI announced last week that developers can now fine-tune GPT-3.5 Turbo and will be able to fine-tune GPT-4 this fall. OpenAI reports that “early testers have reduced prompt size by up to 90% by fine-tuning instructions into the model itself, speeding up each API call and cutting costs.”
The ability to overlay the initial training data with more specific information on a particular industry or topic may be the way of the future for enterprise AI. Coupled with ChatGPT Enterprise, OpenAI is clearly targeting major companies’ needs. The new tier offers extra security, four times longer inputs, sharable chat templates, and unlimited access to the fastest models.
‘Be Flexible, Imaginative, and Brave’: Experts Give Career Advice For an AI World
The onslaught of AI tools can be overwhelming and intimidating. One study shows more than half of workers are using them already. Their promise of efficiency has been well documented — how can you get the most out of them?
Elsewhere …
- Stephen King Shares His Thoughts on AI Writing Fiction
- 31% of Investors Comfortable Trusting AI Financial Advice
- Woman With Paralysis Can Speak By Thinking With a Brain Implant and AI
- Google Debuts a New Way to Watermark AI-Generated Images
Tips and Tricks
Quickly Creating the Perfect Visual
What’s happening: AI tools are evolving quickly — ChatGPT updated earlier this month, and Bing AI, Bard, and Claude all updated in July. Image creation tools have also improved quickly, and it’s easier than ever to create visuals.
Why it matters: The ability to produce images that match written content is as simple as submitting a quick prompt. The results are more consistent than they were a few months ago when, for every lifelike creation — remember the Pope and his puffer coat? — there were 10 images of hands with seven fingers and abstract faces that looked like they belonged in a Picasso.
Try this: The most powerful AI image generator is Midjourney, but it requires a subscription and is only available through Discord. Stable Diffusion is a great alternative and doesn’t even require an account to generate images. You can type in a simple prompt and hit the filter button to determine image style, aspect ratio, and even a negative prompt (what not to include in the image). Since AI art can’t be copyrighted, this is a great way to quickly generate an image that complements written content. As you can see, the Pope recently shed his puffer coat and ran routes at Eagles training camp.
Quote of the Week
“The only thing people forget when they get excited about this is you need to be very specific about what you ask, otherwise it’s like an Excel [s—]-in, [s—]-out spreadsheet.”
— Claus Jepsen, CTO at Unit4 in CIO Dive