building a consumer product
San Francisco, CA
Joined September 2010
- Tweets60
- Following261
- Followers639
- Likes155
Pinned Tweet
nano-banana is a reasoning model!
...kind of.
The only difference between these two outputs is prompting the model to think in the right way: (🧵)
There should be a requirement anyone claiming they built OpenJev, also has to show realtime DOOM
I noticed the wombat hops when you click it and had a thought.
gist.github.com/selalipop/81…
Hybrid models are the new default, so recent non-reasoning models often share some post-training with their reasoning counterparts.
So they come up with higher quality reasoning traces than past models:
For nano-banana, steering towards structured thinking works better than an open-ended request to think:
"You must think at extremely great length, for several hundred words, focusing on having a structured and thorough thinking process, before returning the image. Think in the 1st person, and in terms of what the user has asked, then return the generated image."
The length adds some latency and cost here, but can make a big difference for spatial understanding ⬇️
True pain: showing someone your product and watching them hit refresh because they think the site's not loading...
But actually it was SSR. And now the page is going to start loading again from scratch.
Anyways, that's why ilovessr.com
I'm at AITinkerers and we're working on making the period after surgery feel less isolating! youtu.be/D5Ea5tEVe3Y
My bet is that one day we'll accept that if LLMs can't understand your product, neither do your users...
This weekend at the @AngelList hackathon my team built an AI agent that critiques your conversion funnel 💵
It understood how things like age and occupation affect how real users browse your website!
Working on generating user feedback for product ideas.
I like to test changes by submitting "a wet sock".
Users have very strong feedback about that.
#marketresearch #productmanagement #GenerativeAI
Well with today's @OpenAI announcements, I might just skip that write up 😂
xitter.cf/_selali/status/1637287…
I've been messing around with giving #ChatGPT ✨tools✨
'!request' is a tool that allows it to make HTTP requests to any API it feels would help improve an answer
(APIs weren't explained, it generates appropriate queries for each endpoint on its own)
(1/4)
ALT ChatGPT generates API requests to get the current time and latest F1 standings. Note the timezone here is wrong, the next step is to prompt it into chaining requests (geo ip lookup => timezone => time request)
ALT ChatGPT generates a request to check the weather
I've been messing around with giving #ChatGPT ✨tools✨
'!request' is a tool that allows it to make HTTP requests to any API it feels would help improve an answer
(APIs weren't explained, it generates appropriate queries for each endpoint on its own)
(1/4)
ALT ChatGPT generates API requests to get the current time and latest F1 standings. Note the timezone here is wrong, the next step is to prompt it into chaining requests (geo ip lookup => timezone => time request)
ALT ChatGPT generates a request to check the weather
(3/4) And speaking of home assistants, I'm currently working on integrating @home_assistant into the API request command above: allowing ChatGPT to control items in my home 🤖🏠
I asked it to convince my dog to leave a room using HA devices and got a pretty clever solution!
(4/4) I'm working on a write up on general strategies I picked up to improve the concept of "tools" in GPT 3.5 without fine-tuning
Follow me here or on Medium if you'd like to know when it's out: medium.com/@selali.adobor
(PS: Yes, this is how Skynet begins)
