Видео

Have a look at the latest Stanford NLP Group research that supercedes DSPy: TextGrad. You have very clear presenting style and don't waste time, love it! It'd be great to see a presentation on TextGrad with GraphRAG. Also have a look at this paper, 2407.01502v1 in arxiv, it proves Agentic systems that do a lot of expensive LLM calls are not required.

You rock, this was great!

This is a fantastic video. Where the RAG fails due to potential lost in the middle, is there any trace or context to confirm it was provided to the LLM?

Why is the presenter speaking like a robot?

More like this please!

How does this compare to Graph RAG by Microsoft? What are their differences and similarities?

Thank you! This is exactly what I needed!

LeanChen, if you had to rate the similiarity of groups of words, how would you do it? A bit like are these two classes similar?

fugazi KG- this is only simple LPG (Label Property Graph)

Graphrag is so hot right now.

omg I was trying to hold it together at the introduction when I saw the large soup bowl but I broke down when it said "whatever, y'all get replaced by AI in 5 yrs" 😂 Great video overall. I hope I am right about the size of the soup bowl and you are not 3 feet tall.😅

how does this compare to llamaindex property graph ?

Excellent presentation. Neo4j has done a great job of integrating vector embeddings with knowledge graphs. The neo4j LLM Knowledge Graph Builder for extracting entities and relationships from video transcripts, Wikipedia articles, and pdf format files is impressive. It also merges the extracted knowledge into a graph structure, and then provides an interface to query the LLM about knowledge in the graph.

Thanks for the knowledge (graph) sharing :)

Hey. Thx for the awesome content. Would it be possible to actually show a fully working large scale graph (like proper prod scale thing), and also discuss pros and cons of the approach, and when did you find KGs working well and not so well? The reason I asked is that my own KG experiments worked perfectly for me at smaller scale, but then the speed was so slow, that it was killing my M2 mac all together.

be my teacher,and advisor please connect

Hi it would be great if you could please make longer videos explaining how you did each of these transition for example entity extraction, relationship extraction and so on and then how you did the neo4j integration. Maybe you can make a short video like this out of the original video to attract customers while the long video would still serve as a promising directions for the developer/researcher. I love the output produced from your system, but there’s no way to reproduce what you are doing. Reproducibility is a major concern in KGC.

Awesome indeed 😍Thank you!

Hi, could you convert complex PDF documents (with graphics and tables) into an easily readable text format, such as Markdown? The input file would be a PDF and the output file would be a text file (.txt).

Awesome!

Once the entities are extracted, can the model then be prompted to write a graph query (that could then be executed)? I’m thinking in particular of the “knowing A is B but not B is A” problem, such as when you ask an LLM “who is Mary Pfeiffer’s son?” and it does not say “Tom Cruise” but can answer “who is Tom Cruise’s mother?” just fine?

DiffBot - sound not so good for privacy. That is the role of the DiffBot? Entities extraction? Are there any selfhosted replacements?

Amazing video really. and jokes are definitely necessary. 🤓

too much information on the screen at once ... really painful to follow ... be more sober and straight to the point. Jokes are not necessary

Can't sign in diffbot even with non-gmail ID. Is it supposed to be so?

very well explained, Thanks!

one basic doubt. def forward(self, question): # Step 1: Retrieve context based on the question context = self.retrieve(question).passages # Step 2: Generate an answer based on the context and question prediction = self.generate_answer(context=context, question=question) answer = prediction.answer # Step 3: Validate the answer type using the entity_linker function correct_question_type, original_answer_type, type_status = entity_linker(answer, question) # Optional: You can use the AnswerTypeValidityCheck signature for validation if needed validation = self.check_answer_type(entity_type=correct_question_type, question=question, answer=answer).type_status return dspy.Prediction(context=context, answer=answer, type_status=type_status) in this method, the answer returned is the one you got from self.generate_answer. That method doesnt use any entitylinker . so how is entitylinker influencing the answer ?

I really like this tool can it be used for seo

Do you know of any efforts on converting these entities and relationships further into formal logic representations? Being able to pair these graph databases with formal logic representations would definitely help improve the quality of written text, organic exploration/discovery, and understanding over time.

Do you know of any efforts on converting these entities and relationships further into formal logic representations? Being able to pair these graph databases with formal logic representations would definitely help improve the quality of written text, organic exploration/discovery, and understanding over time.

Your videos are fantastic! Entertaining, educational, and still highly technical.

Do you feel that nomic embeddings are adequate open-source embeddings model for RAG projects or do you recommend another?

I want an LLM that will write a slightly different version of LOTR each time, but without Fellowship

Would love to see this in practice! Seems like it would add a good chunk of latency? I love graphs but they can also "hallucinate" if not given enough training data to build connections right? Or missing data entirely?

Haha. Brilliant. Love the behind-the-scene piece at the end. Very instructive. Thanks 🙏🏻

Here's a ChatGPT summary: - Retrieval bandage generation Bragg is effective in reducing hallucinations in large language models (LLMs). - Despite providing correct context and ground truth, LLMs often do not incorporate external knowledge correctly. - Viewer comment suggests LLMs may prioritize their internal knowledge over external information. - Research compares GPT-4, GPT-3.5, and MST-7B in balancing external information with internal knowledge. - GPT-4 is the most reliable model when using external information, followed by GPT-3.5 and MST-7B. - All models tend to stick to their internal knowledge if they believe external knowledge is less correct. - Rax can enhance accuracy, but its effectiveness depends on the model's confidence and prompting technique. - Different combinations of language models and embedding models can lead to varied results. - The study highlights the influence of different prompting techniques on how LLMs follow external knowledge. - SPI framework for auto-tuning prompts can improve LLMs' adherence to external knowledge. - SPI uses bootstrapping to create and refine examples, improving prompts based on specific metrics. - Entity linking can prevent incorrect answers by mapping and identifying words in text to entities in a knowledge graph. - Default Knowledge Graph is used for validation due to its extensive network of verified information sources. - Entity linker helps filter made-up information when LLMs hallucinate. - DSP RAC pipeline updated with entity type validity check improves output accuracy. - Custom DSP RAC pipeline integrates knowledge graph data to refine questions and retrieve relevant information. - Two metrics for DSP optimizer: entity type check and alignment with knowledge graph context. - Knowledge graph context ensures final answers align with ground truth. - Enhanced output incorporates specific passages and relationships from the knowledge graph. - Example shows knowledge graph confirming Elon Musk as the sole founder of SpaceX. - Optimized program sometimes fails to make LLMs stick to external knowledge. - Manual prompt tweaking may be necessary to ensure LLMs follow external knowledge strictly. - DSPive framework has a steep learning curve but can yield better results for experienced programmers. - Main message: Integrating knowledge graphs and entity linking with LLMs can improve accuracy, but manual prompt customization may still be necessary to ensure adherence to external knowledge.

This is the way...

I haven't seen any good examples of the Self-improving part of DSPy yet. Is it ready for mainstream use❓

thanks great video!

girl, u rock thank you so much for this!! Where can I follow you?

Love this kind of RAG comparison, would be better if the background looks less like a spa room.😅

Until I saw this, I was starting to think that there was something wrong with me not being able to achieve magical improvements in results by using DSPy over meticulously hand-crafted prompts targeted at the observed quirkiness of specific LLMs. Thank you for restoring my self confidence. And now I'm also going to incorporate graph databases into my RAG pipelines after watching a couple of your videos.

great video! thanks

Hey @lckgllm, could you also upload the missing `dspy-requirements-2.txt` in the repo? 🤗

Very interesting, thanks! But Musk wasn't a co-founder of Tesla either.

Thanks 🙏🏽

Great info and content.

Impressive a 996 worker can find time to put this together, keep it up! Ah yeah, RAG doesn't work, in fundamental ways... it can't.

Love it! Finally Neo4j has good usage!

Great video~ Very interested in KG' adaption in LLM. Kindly reminder: hide your api key in the video😉