fbnbr

switching costs are def going down though, so CUDA feels more like the default vs the only option.

real moat is still ops. getting stuff to run is easy, getting it stable at scale isn’t. so yeah, not gone, just more like a tax now.

1. The lab, led by Professor Rodrigo Almeida, is studying economically important grape diseases, focusing on Grapevine leafroll disease and Grapevine red blotch disease [1].

2. They're developing an AI tool for fast and accurate disease identification in vineyards, which could be a game-changer for disease management.

3. Their work combines molecular biology, ecology, and bioinformatics, using advanced techniques like genomics.

4. A recent study led by Kai Blaisdell showed that mealybugs efficiently transmit Grapevine leafroll-associated virus 3 under field conditions, with disease symptoms appearing throughout the plant one year after infection [2].

5. Their focus seems to be more on understanding and managing plant diseases using various molecular and ecological approaches [3].

This research is crucial for the wine industry, especially in regions like Northern California. It's not quite "genomics on the brink of the discovery what was penicillin through crispr", but it's still cutting-edge work that could have significant impacts on grape cultivation and wine production.

References: [1] https://nature.berkeley.edu/almeidalab/research/ [2] https://nature.berkeley.edu/almeidalab/category/lab-news/ [3] https://nature.berkeley.edu/almeidalab/publications/

If you're a government official, advocate, or researcher feeling the impact of this change, consider reaching out to me. I am working on a platform that leverages AI to revolutionize policy research, bill drafting, and legislative analysis. I think there is a lot of toil that can be removed with intelligent policy research, advanced legal analysis, collaborative policy development, and automated legislative drafting. My team and I are designing it to empower teams with tools to facilitate grassroots participation while integrating seamlessly with existing civic platforms.

We're currently looking for early testers to help refine our product. If you're interested in enhancing your productivity and navigating the new legislative landscape with ease, DM me. Your insights could be invaluable in shaping a tool that addresses the real challenges in policy development and advocacy.

What I like about these kinds of solutions is that they address the practical challenges of using multiple LLMs. Rate limits, cost per token, and even just choosing the right model for the job can be a real headache.

KNN-router, for example, lets you define your own logic for routing queries, so you can factor in things like model accuracy, response time, and cost. You can even set up fallback models for when your primary model is unavailable.

It's cool to see these kinds of tools emerging because it shows that people are starting to think seriously about how to build robust, cost-effective LLM pipelines. This is going to be crucial as more and more companies start incorporating LLMs into their products and services.

I think that is one advantage of not just blindly trusting one model but finding consensus among many top rated models within one interface that allows you to quickly cross-check.

I got so excited about reading for Mesos helping in the multi cloud world, potentially as the hypervisor for running k8s

LLMs are nothing alike to compare them with anything close to what’s going on in a brain. It’s all probabilistic and the closest comparison I can think of is comparing a bird with a plane - both fly, but well…

Playground and account are for free

I'm the founder of Pulze.ai, a B2B SaaS Dynamic LLM Automation Platform tailored for developers incorporating AI functionality into their software. We aim to simplify the LLM integration process, letting developers prioritize their core products instead of diving deep into AI specifics.

We've constructed a scoring system for leading models and continually benchmark them. Our platform determines the most suitable LLM to address specific requests based on these benchmarks. To demonstrate this, our playground boasts a compare feature allowing users to share conversational interactions with LLMs—both publicly and privately. As the context changes, we pinpoint various models for responses. These shared conversations can be forked and extended.

Moreover, our extensive API layer isn't restricted to these requests; it encapsulates all the essentials for crafting a successful LLM application. For instance, our logging feature facilitates response ratings, which will soon empower users to fine-tune models, crafting personalized LLMs. These will also be factored into our benchmarks and request routing decisions.

Concerning the comment on LLM benchmarks, I completely concur. Traditional benchmarks or LLM tricks, like acing a particular test, may not be the most robust indicators since they could've been part of the LLM's training set. The genuine challenge lies in evaluating an LLM without compromising the test set and retaining deliberate opaqueness around the questions. Trust issues indeed!

Regarding the Markov chain discussion, I appreciate the insights shared. At Pulze, we recognize the complexities and intricacies of LLMs, and while their foundation might resonate with Markov chains, the scale and depth they operate on are profound.

We've just emerged from stealth, and I'd genuinely value any feedback or thoughts on our approach and platform. Thanks for taking the time!

We tested four different approaches:

1. Classical methods using Term Frequency-Inverse Document Frequency (TF-IDF) with Support Vector Classification (SVC) and Logistic Regression (LG) 2. Sequence classification leveraging BERT and its variants 3. Sentence embeddings with Sentence BERT (SBERT) combined with SVC or LG 4. The SetFit framework that allows end-to-end training of sentence embeddings using Contrastive Learning (CL) and fine-tuning the classification head Our findings suggest that combining pretrained SBERT-based sentence embeddings with SVC provides a robust and efficient solution for prompt classification. Despite the surprisingly effective performance of the classical combination of SVC and TF-IDF, it's suspected to struggle with diverse vocabularies in real-world settings.

The fine-tuned Masked Language Models (MLMs) such as BERT showed mixed performance, with lower suitability for few-shot learning. The SetFit approaches, while promising, appear to need further optimization. We've implemented these findings in the pulze.ai engine with success.

We're planning further research to test these models with diverse real-world datasets and will be sharing more about deploying these models in a production environment in a future post.

We're eager to share these insights with the HackerNews community and look forward to hearing your thoughts and feedback. Please don't hesitate to share your experiences, ideas, or suggestions regarding prompt classification for Language Models.

openai.error.InvalidRequestError: The model: `gpt-4-32k` does not exist

Anyone else seeing this?

It is not listed when retrieving the models via https://api.openai.com/v1/models

But it is listed in their documentation https://platform.openai.com/docs/models/gpt-4

gpt-4 works, just the 32k context seems to be not working. I don't see any official statement about that and am wondering if anyone else is running into this?