rushingcreek

We use Mermaid!

Thank you! Really happy to know that it's working well for you -- we made the additional searches feature exactly for this type of use case.

We used a system of LLM critics to generate a high-quality dataset for this. More in the blog post linked in the answer above.

Thank you! We actually support "zooming into" different parts of an answer. The chat conversation is now stored as a tree, so you can ask a followup, then revise the followup, and switch between the revisions below the answer when doing a third followup.

Interesting, do you mind sharing some of the queries you tried? We avoid doing automatic searches for certain followups if the existing sources already contain the correct information, but we should be giving improved sources compared to the original version of Phind.

We have a products UI on the way :)

With this release we are aiming to be simultaneously better for developer questions while also competing as a broad AI search engine for just about any question.

Thanks for the feedback. We're working on making the answers a lot faster for the Phind models. I can see how it would be frustrating to be waiting for the diagram to generate when you're waiting for text or for an example.

Thank you! We went quiet while we rebuilt the entire product, but now we're back and will do everything we can to stay at the forefront.

Ah, great question! We tried using off-the-shelf models and found that they are incredibly bad at generating these visual components reliably, let alone with the nuance (such as color, placement, details, etc.) that would be expected in a great answer.

We write more about this in the technical model blog post: https://www.phind.com/blog/phind-2-model-creation.

Thank you :)

Thank you!

Thank you! We do plan to support an API this year. We have deprecated our VS Code extension, however, as we're going all-in on search.

We'd love to partner directly with content creators. This is something we're thinking about a lot and are having conversations on how to go about it.

I will also personally do everything in my power to keep us from running ads on Phind. I want to keep Phind honest and authentic, and will do everything I can to make it a net positive for the internet.

Thank you! Appreciate the feedback. Custom links are something we should be able to do -- we'll run some checks and make sure that feature is working properly. Thanks for letting me know.

Thank you :)

Yep, we trained the model to produce Mermaid when a diagram might be helpful. There are a lot of nuances that went into this, such as where in the answer to place the diagram and how to ensure a good contrast within the diagram for maximum readability.

We have some more details on this in the model technical deep dive blog post :)

Our new 70B is actually completely different from the one we launched a year ago -- we should probably update the naming to make that more clear!

What do you think about its performance on your queries?

Thanks for the feedback. Have you tried setting your answer profile in https://www.phind.com/settings/profile?

You can tell it to "only answer using plaintext" there and it will be automatically applied across your searches.

Would love to see how DeepSeek R1 compares to O1 here.

This is an awesome demo and I hope you succeed.

I've dabbled in building a tool of this type and something I've learned is that making single-task demo apps (like chess or snake game) is much faster and simpler than making production-grade apps that could make a real business. In fact, the process of making these demo apps production ready (polishing the UI, adding features, etc.) would usually result in the user giving up. In our implementation, it was much trickier for the system to understand the nuances of polishing than getting the broad strokes of the initial demo. I'd love to know if you've ran into this and what your thoughts are on getting around it.

As there was no mention of an API for either Sora or o1 Pro, I think this launch further marks OpenAI’s transition from an infrastructure company to a product company.

Thanks for the detailed feedback, this is super helpful. We’re working on a complete redesign at the moment to fix the longstanding UX issues along with a suite of model updates. I’ll be in touch over email — I’d love to give you a preview and get your thoughts.

Thanks for the feedback —- can you share a bit more about what you feel has gotten worse? Have you just been using the default model?

We're working on a complete redesign of Phind to make it more of a search competitor :)

What were the limitations you ran into?

Great feedback!

Thanks! Have you tried enabling “multi-query” mode in the search box?

That’s great to hear! What are you asking it?

We run the nodes "hot" and close to overload for peak throughput. That's why NVIDIA's XQA innovation was so interesting, because it allows for much higher throughput for a given latency budget: https://github.com/NVIDIA/TensorRT-LLM/blob/main/docs/source....

Serverless would make more sense if we had a significant underutilization problem.

The "for everyone" part is about the new Phind Instant, trained using similar data to Phind-405B, which is great at fast information retrieval