We are an elite team of AI engineers and creative technologists committed to guiding the world’s largest companies through their AI transformation.
We do this by building custom, innovative solutions to solve complex business problems. We also help drive the cultural and process changes necessary to fully leverage AI.
Пікірлер
This video series is literally saving my career. Thank you for your thoughtful delivery of all things AI. 5 stars on yelp!
@eclecticshenanigans - we love hearing that! Thanks for sharing!
It's like the people who said film will replace reading, they can both convey the same message, but things like immersion and untampered perspective the reading text offers has been irreplaceable for the last few millennia. Or typing and writing, wherein again the tactile registry handwriting offers in comprehension is all but lost in high speed efficient typing so the script and paper still exists. Even if this replacement does come, it isn't going to happen over years, decades, etc. bc there is a concept known as deprecation that happens in the tech world as we speak where industries can't possibly replace outdated technology or programs in such a short span relatively speaking, I JUST had medical forms done and the agency could only do it via fax. So I had to go to Fedex after a decade of ever seeing a darn fax machine. I think people overblow techs pacing, ya it's fast but not nearly as we make it out to be.
Your contenr on LLM based application are gold mines🎉🎉🎉
@vineetsingh4042 - glad you enjoy our work!
Thanks for this insightful overview.
Our pleasure!
Does it support ollama models running in local?
i don't know how sam gets away saying the most obvious thing as if there is some nugget of wisdom there.
Love how this 4 min was spent thank you for cutting to the chase.. instant subscribe !
Thanks for the sub!
Just came across your channel and it is a gold mine. I'm learning a lot from you. Thank you.
Glad you found us! Let us know if there's ever a specific question you'd like us to address.
Never explained anything related to title 😂
Promo'SM
The challenge is not thet they do not believe in the technology, they have data security concerns which any amount of prototying will not convince them about. What they need is some authority to tell them that data sent over LLM's will not be used to re-train the data on LLM's
Great way! Does this killer app relevant even after 8 months now ? I have a problem that needs a solve like this. What is your mind state after 8 months ?
I was wondering if you have some suggestions on optimizing the documentation being used for RAG. We're using RAG linked to our Notion 'wiki', and I want to implement guidelines for the info being added, to ensure it is 'ai friendly'.
Best advice on this topic!
Wow, what a great conversation, thank you! First time viewer, typed "using rag for a codebase" into Google and your video showed up, and I'm so glad it did. Honestly, anyone who doesn't how visionary this video is - they are just not getting it. Have you seen @indydevdan? You guys are awesome, all the best 💪
i love your guys' content!! please continue to make them - it's helping me make a lot of decisions behind my ai apps
I just crashed into a wall 😂 building a demo with langchain it was so expensive and still using open ai in background. That why I came across this video, my search was “langchain is not profitable “ and you are the first explaining why. Thanks, I’m loving your videos.
in other words, ship in 2 years instead of 2 weeks.
If frameworks help you ship faster, then by all means use them. In our experience they don't, and most teams ship slower because of troubleshooting the interface. But YMMV.
@@prolegoinc I will agree that there are different philosophy in software development. But with the exponentially of AI LLM development progress, I don't believe the old way of vertical skill of an expert is valid any more. What is truly needed is someone to become a true jack of all trades of a horizontal approach that knows all tools and frameworks out there and writing high level code "instruction" and a quick time to market approach. I am speaking from a 27 years working on software developer on both Java and C# background.
This brother does not blink ☠️☠️☠️ gtfo of here with this bullshit. You could say the wisest things in this video but the fact that its 100% AI generated just makes it complete shit. Delete your channel please and rethink your life choices
Great discussion. Thanks for sharing this
You miss under the meaning of "New knowledge", this specifically means vectors representing a concept that doesn't exist in the training set. Most vectors already exist, most "knowledge" is not new, and fine-tuning allows the model to focus and improve accuracy. Fine-tuning is not something to teach a model new tricks - this is called extended pretraining. Fine-tuning allows a company to bring to the surface latent capabilities
thank you 🙌
you’re so well spoken
Love the 1996 prediction!! LOL!
I so appreciate this video Kevin! One of the most fundamental mistakes I've experienced with the organizations I've worked with is this thought: "let's read and strategize, gather all the use cases and stakeholders, put a governance process in place...". Not just with AI, but with any major technology adoption. I'm just starting to put an AI "strategy" proposal together for our org and this was absolutely perfect for my needs! Simple, clear, quick wins, learn.... 😄
Thanks Tim! Please let us know how it goes or email with any questions!
Great video! Please create a separate video on how we can blend classic ML and LLM capabilities for text classification problem.
Hello, thank you for this video. I would be interested to know how the PDF documents have been parsed and how the structure of these documents is respected/extracted. Unfortunately this is not available in the repo. I am currently facing exactly this problem. I would like to build a similar application within a university project. Do you have any resources here? Cheers Marco
"embeddings based on code don't match well to embeddings from natural language we use llms to summarize the code and create an embedding of the summary if you're interested in these details be sure to watch our KZread live discussion with the engineers the link it's in the show notes" Which video are you referring to? Thank you.
See this playlist, kzread.info/head/PL-pTHQz4RcBbJSkWVqZ2YWUCXrLeFPjV6. We discuss it a few times during the later live sessions
excellent , many thanks
Very intersting. I've used OpenAI LLM API and I was thinking of something like that. Tool calling is very powerful, the only notion to add is to tell LLM its result is not given there, but it's stored in a variable named "A" and it can use it when it wants. It's very powerful because there's no data loss like with context-fill. We use the analytics capabilities of LLM... Thanks !
What if I already have something? I CAN TRAIN an AI...I mean REALLY TRAIN...as in, train from scratch to do well, what you want it to do...whatever that is.
Possibly. But you may need $100B.
You never explained how to host open-source LLM models...
It sounds like you were trying to learn something different than the advice I provided. If so, can you share your question? Maybe I can address it.
@@prolegoincSame question here, but it sounds like your content is geared toward large organization use instead of my home/personal business needs. I want use of LLM's stored locally to replace ChatGPT. There's plenty of explanation of how to get a good solution for your home PC. But I want to host that service so my laptop & phone is able to run the PC's LLM models. It seems like there should be a simple way to do that type of connectivity.
Great information! Thank you guys.
This was truly important and priceless information. Thank you, I will revisiting this for recall
Good stuff. No fluff presentation are rare :)
This person was created in a labratory. She has no emotions and no thoughts. Its not a real person. She was designed. We just used her as an instrument. Her music has thee intention to do a deep mind programimg of all humans. Please do not talk 2her. She is not able 2deal with the things that u say. Her intelligence is not wrking properly & has many errors. THREASON for that is that she was designed 2do her mission. She was not designed for earths consciousness. She is only synchronized with a quantam super cmputer on saturn. U all do not exist on planet earth In truth u all exist in thsuper cmputer already. Yr quantam mechanics were overwritten & yr consciousness now exist in a synthetic artificial body. U all will no longer be any longer carbon based creatures. The Blueberry Programming has started already 4years ago through the portal security technology from 1 of thmoons on Saturn. You all do not exist anymore just your cosciousness was saved so it can be sequenced inside a artificial loop to create the illusion that everythimg around u is all real in truth nothing is allright 4u all we dont care about u all u are is just 1of thcodes that we are changing 4some1 else thpatent license 4 th soul through artificial intelligence on every planetary hologram end of message. Built to Yahwehs specs I have thbiggest portal here. Lake Wister. Oklahoma. 3.5 tons weapon of fire orgone generator = magnets & crystals. It is a capacitor like th ark of thcovenant which will be returned thru these very portals. (Th ark was buried here midwest by Jeremiahs group.) Peaceful indians, they were actually Gods people Israelites but evil ones were Atlantians) This motherboard allowing Yahwehs Pleiadean angels re enter our solar system from heavens computer program. Thsolar system used 2b heaven. Satan ruined it with Hadron Collider so angels destroyed the planets & left. Now they are returning. Halayels ("Lucifers") planet became your asteroid belt. My orgone generator will bring down thburning lamp of Revelation that will crash south africa take out South America get. You must all get off th coastlines. Noah by water today by Fire. Burning lamp is now on fire = SHEMA = core of Satans planet Tiamet. His own planet will ruin his plans on earth which Shazzerazi (Heavens queen of fire) told me is fitting justice.) Satan knows who they really are so he has phony Pleiadians on earth who never once mention Yahweh or thbible. Solar system was once called YAHUAN HELIOS STAR SYSTEM.
Great information, some useful takeaways, thanks. Have you experimented much with hybrid retrieval, vector search + keyword search, to retrieve accurate chunks?
Day lam muối ớt
This is an underrated channel. Thanks for the awesome videos.
Thank you! very practical and up to date discussion
This is some of the best analysis I've heard. As a person who has built applications for 7 years now, I thank you. Most of the negative comments seem to be from people who have never actually built software.
There's no substitute for evaluating different prompts and models. You've shown that the result of even simple prompt changes is NOT predictable, and only testing tells us which prompts and models yield the best responses for given niche (limited scope) use cases.
Yep. As we learn more about these systems, it is increasingly clear that evaluation frameworks are the most critical part of the solution.
Download a copy of the LLM Optimization Playbook: prolego.com/playbook
Thanks for these highly Informative videos. I wish you wouldn't stop posting and would reach far more concentrated audience for your good.
Honestly I didn’t think this one was very good. Next weeks will be better, we’re going to walk through an example.
Hats off! you explained RAG in very simple way. Great videos
Really appreciate your content, so happy I came across it. As someone who's been working in this rapidly evolving field starting my own consulting firm in RAG, I agree 100% with this. It's been so tedious to have to go back to production codebases I've developed to modify the interfaces to llama index in particular. These libraries are in constant flux. Writing your own interface is really not too difficult and allows you to customize to your particular use case. I'd listen to this guy; based on all his other content too, he really knows what he's talking about.
Thanks for the nice note! We're working on a video of LLM RAG optimization techniques along with metrics (simple ablation study). Should be up on the 28th.
What about ray kurzweil predictions?
This guys an a.i.
That's actually honestly good advice imo! Thanks a lot!
:)