Ada Boese

Learning about LLMs as a I build the best article generator the world has ever seen!

🟩🟨⬜️⬜️⬜️⬜️⬜️⬜️⬜️⬜️ $50k/m

[Ada Boese @AdaBoese]

In early days of developing SaaS, your focus has to be developing relationship with users, not revenue maximization. Otherwise you will miss out on tremendously valuable feedback.

[Charly Wargnier @DataChaz]

Must bookmark.

@PunkpEye has put together a huge collection of 300+ MCP server implementations tailored for AI agents! 🤯

This Github repo links AI models to real-world tools such as web browsers, APIs, databases, cloud services, CLIs, & much more!

↳ https://t.co/CX3YywWXFx

[Ada Boese @AdaBoese]

@AlexBelogubov It is a tough space/tough problem. I've been working on AIMD app for several months now. Focused on quality beyond anything else. I have good results, but the application got extremely complex to maintain.

[Ada Boese @AdaBoese]

@shwnhll @carriganisms I should give this a try

[Ada Boese @AdaBoese]

@adcrookes I think we should partner!

[Ada Boese @AdaBoese]

@hiteshjoshi @ycombinator You make it!

[Ada Boese @AdaBoese]

@phuclh93 https://t.co/cuorzDPQw6 integration?

[Ada Boese @AdaBoese]

I don't know what changed, but AI generated content is taking off the ground big time this month. I am increasingly using https://t.co/ql0f4iGwb8 for my own projects rather than focusing on selling it... and holy guacamoly did I see a shift in Google Search Console this month compared to previous.

Google algo changes?

[Ada Boese @AdaBoese]

@pbteja1998 So much love for this project! Been checking the progress since the start. Amazing how much you've achieved!

[Ada Boese @AdaBoese]

@AlexBelogubov @lilybankai Still very much not working

[Ada Boese @AdaBoese]

I've been very quiet. I don't know if it is good for my social presence, but it is been very good for my product growth. Need to find a balance of the two.

[Ada Boese @AdaBoese]

@AndrewPierno Send

[Ada Boese @AdaBoese]

Everyone keeps throwing random suggestions at Google. "Google should do X, Y, Z for the best results"

It is not hard to find good results in a small datasets. Everyone seem to ignore that whatever they do must be feasible with trillions of datapoints

[Ada Boese @AdaBoese]

What happened to all the speculations about GPT5 coming out?

[Ada Boese @AdaBoese]

@SaidAitmbarek Thank you for the shoutout!

[Ada Boese @AdaBoese]

This experiment has been pretty nuts

[Ada Boese @AdaBoese]

Claude vs. OpenAI GPT-4 generated content, side-by-side comparison

- OpenAI GPT-4 https://t.co/jWWUN0hDCe
- Claude 3 (Opus) https://t.co/EV5EiV1uJj

Both of these are outputs of AIMD app. They are not made using a single prompt, but rather using RAG with over a dozen instructions. This allows to test a quite broad range of expectations, such as the adherence to instructions, error rate, speed, etc. Since the two model APIs are mostly compatible, I've decided to compare it side-by-side.

A few interesting observations:

- Claude followed instructions a lot closer than OpenAI. The outline that was provided to the initial instructions is pretty close to the final article structure despite multiple revisions.

- Claude output scored better in terms of use of broader set of data formats (tables, lists, quotes).

- Contrary to many tweets, Claude output is not excessively verbose. Worth mentioning that part of RAG instructions to rewrite content for brevity.

- Claude took 5 minutes to execute 52 prompts. OpenAI took 7 minutes.

– Image inputs (prompts) are generated with respective models, but the actual images are generated using DALLE.

[Ada Boese @AdaBoese]

@mattshumer_ @levelsio This works equally well with OpenAI just in case. That's the approach that I use.

[Ada Boese @AdaBoese]

@levelsio @AnthropicAI I've not seen anyone talk about throughput speeds. How are those? thinking of trying this out for https://t.co/t5lqyJirFm, although pretty happy with current LLMs

[Ada Boese @AdaBoese]

https://t.co/t5lqyJhTPO is by far the most advanced article generator, using several different LLMs and external APIs to create content that's both easy to read and contains rich data with images, quotations, tables, etc.

[Ada Boese @AdaBoese]

@illyism @ProductHunt Congrats!