Tom Blair

Large language models are wonderful at ingesting large amounts of content & summarizing. Uploading an academic paper and I can pester it with an infinite list of questions & it will respond with equally infinite patience. In comparing the two most recent Microsoft earnings calls, Claude highlighted: - faster than expected Azure growth (29% vs 27%) - AI contributing 8% of Azure revenue up from 7% - higher CapEx spending & greater capacity constraints for data centers - much better commercial bookings growth : 17% vs 13% Excellent analysis. Benn Stancil described LLMs as great averagers of information. But what if I don’t want the average? If I seek the data point two standard deviations out? If I’m channeling my inner Anthony Bourdain & I’m seeking the fermented shark or cobra heart within my query? At Google, we ranked web pages & ads with many signals. An engineer taught me that of them found rarity of a word across a set of documents. For example, if there’s a document containing 10 instances of the word ghotiacross a collection of documents, where the average document has 0 ghotis, that document is likely the best to answer a search about fish. I haven’t found a way to goad an LLM to produce the rare result. Why did one user decide to use a particular piece of software when 10 others did not? Maybe I haven’t learned how to prompt an LLM well. Ideogram launched a feature called Magic Prompt that expands a basic prompt into richer instructions that elicit better results, marrying the language of a user & a computer. A user might write : A boy with a dog in a park. Magic Prompt replies : A heartwarming scene of a young boy playing with his happy, wagging dog in a lush, green park. The boy has a playful smile on his face while he tosses a bright yellow ball for his furry friend to fetch. The park is filled with beautiful flowers and trees, creating a serene atmosphere for this bonding moment between boy and dog. Magic Prompt surprised me in three ways. First, unlike Google search where terse queries tend to work very well, LLMs relish verbosity. Less Hemingway & more Faulkner. Second, maybe there’s a prompt for rarity. If you know what it is, please share with the rarity shibboleth. Third, I wonder what other prompt techniques are out there that have yet to be discovered to wrest the best from an LLM. This technique is called TF-IDF. It means Term Frequency Inverse Document Frequency.

20

5 Comments

All of these statements about enterprise AI could simultaneously be true, even if they seem somewhat contradictory: 🌟 Hype will come and go, but GenAI will be a game changer for automating mundane parts of numerous enterprise workflows 🏗️ Making GenAI work within core workflows in the enterprise takes tremendous hard work beyond getting the tech to work - meticulous data pipelines, ensuring accuracy across real-life scenarios, safeguards against hallucination, robust guardrails, on-prem deployment capabilities where needed, change management and more, all while ensuring predictable and acceptable costs. Once done successfully, all this effort creates moats ⚙️ Large models may work out-of-the-box for some basic horizontal business tasks. Applications built around domain-specific models will work much better for many other tasks. Agentic workflows are one likely path to broader automation 👥🤖 Human Augmentation and Human-in-the-loop automation is what is here-and-now. Full automation solutions - outside of simple, low consequence tasks - will take time to perfect and be sufficiently trusted in the enterprise 🚫🎯 Enterprises will not buy “GenAI”. They will buy solutions and outcomes. If a GenAI-powered solution is the optimal one for their needs, then that will be bought 🔄 GenAI has opened tremendous opportunities for enterprise software startups that do other adjacent things to make their products more relevant, usable or actionable, and ‘close the loop’ ⚠️ Many “GenAI for X” startups will fail to deliver value in the enterprise. Many GenAI-native startups with the right solution and approach will be massive home runs. Success comes from focusing on the entire solution you are delivering, not just the technology that enabled it 🛠️ Some Enterprise AI startups will need to become services companies in order to deliver value 🕶️ Some Enterprise AI startups will succeed without becoming services companies 🤹 Both open source and proprietary foundation models will be used in enterprise AI deployments - several models may get used within the same application. The best model will win - on a per API call basis For many things being talked about enterprise AI, the claim and counterclaim could both be simultaneously true. Augmentation and automation of enterprise workflows is a massive undertaking that is broad, wide, deep and extremely bespoke. AI-fication of work and workflows is one of the biggest opportunities in our lifetimes. Things are just getting started!

48

10 Comments

Influencing Large Language Models (LLMs) or Page Rank on Steroids In a previous post https://lnkd.in/gMbBP_fZ, I discussed how Large Language Models (LLMs) utilize open-source data to generate weighted responses. Let's go a bit deeper into how LLMs incorporate data and how you can strategically influence their future outputs. While you read think hard about the implications. Recently, I engaged directly with a popular LLM to gain insights into its data integration process. Below is a summary of what I learned: Key Factors Affecting Data Incorporation Model Updates: "We periodically update models using publicly available training data, which is released in specific cycles." Explanation: LLMs are periodically retrained using publicly available data. The inclusion of new information depends on when these data sources are refreshed and ingested by the model. Content Popularity: "Frequently shared information across reliable sources increases its importance and likelihood of incorporation." Explanation: If information is widely shared across multiple trusted platforms, it becomes more likely to be incorporated into future LLM updates. Update Frequency: Models are updated depending on the release schedule. This means any new content or data that is widely shared or referenced may happen immediately or take several months to be integrated into future versions. This is controlled by the developer team and business executives as they see fit. Content Influence: Well-posted and frequently cited information, especially on reputable platforms, has a higher chance of being reflected in future LLM responses after the next update cycle. Amplifying Your Impact on LLMs To increase the likelihood of your content influencing LLM outputs, consider the following strategies shared directly by the LLM: Distribute Content Broadly: "Share your content widely to influence trends and future updates." Actionable Tip: Maximize the distribution of your content across various platforms and communities. The more frequently your information is shared, the more likely it is to be picked up by LLMs. Leverage Reputable Sources: "Contribute to established platforms, such as technical forums, trusted blogs, and community knowledge bases." Actionable Tip: Focus on contributing to highly credible and well-trafficked sites. Some examples include: Technical forums (e.g., Stack Overflow) Reputable blogs in your industry Community-driven knowledge bases (e.g., GitHub, Wikipedia) By strategically shaping what is defined as high-quality information across trusted sources, you can significantly enhance its chances of being included in the training data for future LLM models. This approach is akin to influencing "Page Rank on steroids," where the more seemingly credible your content is and the broader its reach, the more likely it is to shape future LLM responses. #ai #pagerank #artificialwisdom #precision_vs_accuracy

2

#neurips #realtime #insurance #ai First up Waymo By Google. 📡🚗🤼‍♂️ - 150,000 rides/week - no humans in the loop - minor risks - new cars coming. - lidar, radar, vision, and audio capture too - steering. and acceleration. - all about sensing and understanding of the word. - pedestrians are real and unpredictable - night... There is risk and insurance that is special will be required. What's the problem? "the real world is very unforgiving .. many agents on the road which don't follow any rules" At Google Waymo AI is used for... - learn from rich sensors in real time - act and react , prediction of the environment - A big challenge is generalization - the driving stack is sensors, perception, prediction, planning and controls - maps and predicted maps - plan validation - data compression - seeking an efficient transformer for birdseye view processing (beyond ConvNets) - long view birds eye... real time - behavior modeling and prediction with gtp's, trajectories as sentences in a LLM, vocabulary is state and motion, local continuity and global context ... motion token sequences - joint prediction with temporal causal relationships - now using diffusion models to (new) gen scenes and predict world playing forward Scaling ? - architecture for scales in implicit in design - the laws of scaling are not the same as one LLMs (order of magnitude) - inference and compute timing is important - These are TeraFLOP models in real-time - Size matters, bigger better for now - Auto labeling helps eg with 3D bounding boxes - MoDAR is a new thing - motion forecasting ..supports maintenance of a word view beyond lidar and visual. consistency of motion - closed loop training and reinforcement learning, triand on human driving data - bringing VLMs and LLMs into the picture.. like traffic sign understanding, layer in multi modal models for zero shot labelling , motion prediction ("works amazingly well") - exploring CoT reasoning while driving, meta-decisioning and reflexion - multitask 3D reasoning ... with positive interference - combining models on to a "knowledge of the world for AVs They will need Insurance experts.

2

Modern GTM Data should look like modern data for LLMs Harry Stebbings episode with Alexandr Wang a few weeks ago smacked us in the face with this. Alex is the leader in the #AI data space. He said frontier data is what is going to power these models going forward. New data, complex data and data that is created by leveraging our intellectual capabilities in reverse engineering how we as humans would interact with models.  Chain-of-thought prompting is a prompt engineering technique that aims to improve language models' performance on tasks requiring logic, calculation and decision-making by structuring the input prompt in a way that mimics human reasoning. This is required after you apply the reasoning that enables AI models to break down complicated problems into easy-to-handle intermediate steps, leading them to solve complex multi-step reasoning problems. Wait? Didn't I just do this in Clay in a demo two weeks ago? Isn't this all the rage in GTM data? YES - it is. GTM Architecture isn't changing - it has changed. The architect is the most important player right now. We need to reverse engineer how to be successful with GTM data which means we get meetings, we provide value and we get a mutual exchange with our target audience in a consistent and repeatable manner.  You need the basic building blocks to chain these thoughts together: 𝐃𝐚𝐭𝐚: Can't just use the same old data providers and win. Dozens of data sources need to be combined, enriched, verified and 𝐚𝐧𝐚𝐥𝐲𝐳𝐞𝐝 𝐭𝐨𝐠𝐞𝐭𝐡𝐞𝐫. 𝐄𝐱𝐞𝐜𝐮𝐭𝐢𝐨𝐧: This can all be automated. No need to pay someone for outcomes you expect from them when they have to design this manually and at best will take them 40% of their time but they have no experience in this. It isn't their fault! 𝐈𝐧𝐩𝐮𝐭: What do you do to the data that makes it more meaningful? Hiring, spending, technology, patterns, competitive information.  So much more. 𝐀𝐈: Yes. We need #AI and it should be used. However, prompting is a skill and needs to be done right and yes - humans need to be in the loop 𝐄𝐦𝐚𝐢𝐥 𝐒𝐞𝐪𝐮𝐞𝐧𝐜𝐞: You can't rely on the old way of delivering. Need multiple domains, need them rotating, need verified emails and you need a system that runs lights out that you don't support 𝐓𝐫𝐢𝐠𝐠𝐞𝐫: When is the best time to fire my cannon? 𝐑𝐞𝐥𝐞𝐯𝐚𝐧𝐜𝐲: Why does this even matter, at what point, what is the value? 𝐕𝐨𝐥𝐮𝐦𝐞: Less will be more. Shotgun approach and spray and pray looking at open rates and applying the wrong math won't work. A lot has changed. The challenge is fun. Data used in AI systems can teach us GTM folks a lot. When patterns tell you what is going to happen - listen to them.

22

1 Comment

Large LLM API providers are rapidly lowering the cost of using their LLMs. In the absence of a clear leader in model quality, price has become the main competitive factor. Also most likely they see much larger revenue potential from B2B market and integration of LLMs. 🛠️ On August 12th, Google will drop the prices of Gemini Flash by 80%. This seems to be a response to the GPT-4o mini, just two weeks after its release. It would be great if they did the same with Pro 1.5. 🛠️ OpenAI The new version of GPT-4o has seen its price drop to $2.5 input/$10 output per million tokens, compared to the previous versions, which were $5 input/$15 output per million tokens. Over the past month, Claude 3.5 Sonnet and Llama 3.1 405B have significantly pressured OpenAI on pricing, forcing a response. 🛠️ DeepSeek is releasing the Context Caching they promised a month ago. The prices are astonishing: the cost of tokens when hitting the cache drops not by 2x, like Google’s, but by 10x, with free storage included. Also Google recently announced not only a price drop for Gemini Flash but also made model tuning available to all developers. I missed this announcement earlier, but it’s a game-changer: - Tuning is entirely free—not just for Flash, but also for Pro. - The inference cost for tuned models is the same as for the standard models (unlike OpenAI, where tuned models cost twice as much). This likely works through a method similar to batched LoRa, which means the quality of tuning on large datasets might be affected. However, this needs to be tested to understand the impact. Additionally, the Gemini API can be used for free within certain limits, including the ability to tune models based on your own requests. Here are some relevant hashtags: #AI #LLM #MachineLearning #APIs #GoogleGemini #OpenAI #GPT4 #DeepLearning #TechNews #AITrends #AIInnovation #ArtificialIntelligence

10

Friends, it's time... We're going to dive 🏊🏽‍♂️ into the Mathematical Model of Revenue Architecture...👨🏽‍🔬 Because none of this matters, if you can't count ➕ ➖ ✖ ➗ or know of these effects 📈 on which you build a SaaS or AIaaS business. Let alone that you will T3D2 🚀 . Unless you leverage and weave these principles and mathematics into everything you do in terms of GTM 👨🏽‍🚀 . T3D2 is a concept in our business where as soon as you hit 1 mio ARR, you triple (3x) every year and then you double (2x) your revenue for two years, bringing you to a theoretical 100 mio+ ARR (and IPO). Today on average it takes 14 years to IPO, but that's a different topic. We're here for the math today. My math teachers made 1 critical mistake when trying to educate me on these principles: they expressed it in theory - not in money. 🐶 Jacco van der Kooij taught me many things, like reading again, and much more often. But one of the most important gifts was falling in love with math. But a different kind of math than you're taught at school: GTM math. Like how hypergrowth is the result of non-linear behavior and things like COMPOUND and EXPONENTIAL impact... It's a bit like magic - but then science magic, if you will... And perhaps most importantly: how far that REALLY goes or is in every little thing you do in your GTM system, aside from the data models itself. Math matters. And you better get your math straight. I'll help you with that. This week, I'll cover: ⏺ The elements of Accelerated growth ⏺ Linear vs. Exponential Impact ⏺ Nonlinearity in Acquisition ⏺ Nonlinearity in Retention & Expansion Looking most forward to taking my famous lab coat on and diving into the theory with you folks. School tomorrow starts at 2.06 pm. Don't be late.

27

17 Comments

o1 is a new dimension for scaling intelligence. Architectures like this will increase the pace of AI development but also obfuscate what's happening inside them. I expect we will start seeing more similar architecture emerge from open-source models and packaging them into a single API call. BTW you can get similar results with architectures like Mixture of Agents (https://lnkd.in/d6nxZ4NE) or agent orchestration now, o1 is of course more optimized and faster. Previously, scaling of AI systems was about how much data you put in and considered. The time it took and the quality of the response was relative to that. o1 introduces time as a dimension that allows for more considered responses. It was already true that they might not be great at consistently producing good results, but they are great at checking whether a result is good. Pic: OpenAI's explanation from the docs

12

2 Comments

🚀🌊 The Future of Data Exploration 🌊🤖 Imagine standing at the edge of a vast ocean, knowing there’s treasure beneath the waves—but all you have is a rusty snorkel. That’s how traditional data exploration feels in 2024. Now, picture a sleek submarine outfitted with AI sonar, cutting through the depths, uncovering insights you didn’t even know existed. That’s the vision in this week’s article from AI of the Coast: a deep dive into how cutting-edge AI tools are transforming our ability to make sense of complex data ecosystems. For fellow founders and data enthusiasts, this piece will make you rethink how your teams approach discovery and decision-making. The future isn't just about finding answers—it’s about asking better questions, faster. Take a look, and let me know: are you still snorkeling, or have you found your AI submarine? 🌊💡 #Startups #AI #DataExploration #Innovation

71

109 Comments

#ObsessivelyEnthusiasticInvestor #UnreasonablyEnthusiasticInvestor #IrrationallyEnthusiasticInvestor at REFASHIOND Ventures: The Industrial Transformation Fund. We look for the people we believe are doing the best work in Industrial Transformation & Supply Chain Technology; Attempting the impossible. Then we go all-in on them. With obsessive and unreasonable enthusiasm. Join us. #IndustrialTransformation #SupplyChain #Innovation #Technology #Startups #VentureCapital #DemandTheImpossible #UnreasonableEnthusiasm #ObsessiveEnthusiasm #IrrationalEnthusiasm 

6

All my recent dreams for DuckDB came true with version 1.1. The GeoParquet support is awesome. DuckDB was already my go to tool for working with GeoParquet, but I always had to tell DuckDB to parse the WKB, and then convert it after I exported it out. Now everything 'just works'. And then the bonus is that they now support Brotli for compression. This is definitely the smallest compression option for geospatial data, and our data is often huge, so being able to read and parse it means that I now feel comfortable using it. If you're new to DuckDB and/or using it with spatial data check out the post I wrote a year ago when I first discovered it: https://lnkd.in/dkp7GERu It's remained at the top of my list of tools, indeed just now I was using it to help me experiments to get Planetscope data into STAC GeoParquet.

97

2 Comments

🆕 Benchmarks 201 with Clémentine Fourrier of the Hugging Face Open LLM Leaderboard! Why Chat Arenas are **OVERRATED**: - they bias sycophancy/assertiveness over factuality - annotators are not representative of humanity - models are argmaxxing public arena data - not reproducible - can't give fine grained test of model capabilities Why you should STOP using LLMs as Judges: - Position bias (prefer first answers) - Self bias (prefer their own answers) - Length bias (prefer long answers) - Discrete bias (can't do continuous ranges) - Reproducibility (GPT4 closed source - prefer Seungone Kim's Prometheus or Beijing Academy of Artificial Intelligence(BAAI) JudgeLM) One of our best ones 🔥

30

1 Comment

Yet another sleepless night catching up on GenAI research papers, and my mind is blown! My co-founder Matthew Wallace came across a paper suggesting that LLMs can scale to billions of parameters without using MatMul. arxiv.org/abs/2406.02528 Further it says due to the scaling properties shown MatMul-FREE models would actually out perform current model architectures as they scale even larger. This breakthrough, once proven and adopted, could lead to human-level brain efficiency for AGI. Let’s dive in: Implications: 1. GPUs in their current usage are heavily under-optimized. This breakthrough could lead to a massive increase in their CURRENT capabilities when optimized. Jevon's Paradox !!!! as the increase in efficiency would just lead to more and more usage of current and new GPUs to produce more and more LLM capabilities well ahead of current predictions. 2. CPUs and FPGAs could become far more efficient in training LLMs and even more efficient in inferencing them. Imagine your personal AGI on a USB stick! 3. AMD (Xilinx) and Intel (Altera) can potentially counter the NVDA competition with FPGAs. 4. True real-world real-time inferencing at the edge for nearly any use case becomes possible. First, what is MatMul? MatMul, or matrix multiplication, involves multiplying large matrices, which dominates memory usage and computational resources. This process is why such large clusters of GPUs and extensive memory are needed for training modern LLMs. Reducing or eliminating MatMul could drastically cut down on the hardware requirements and costs associated with training these models. - By using an optimized kernel during inference, memory consumption can be reduced by more than 10× compared to unoptimized models. 🤯 - MatMul-free LLM: Replaces MatMul operations in dense layers with ternary accumulations using weights constrained to {-1, 0, +1}. This reduces computational cost and memory usage while keeping the network effective. - Self-Attention Optimization: Uses a modified GRU, the MatMul-free Linear GRU (MLGRU), which relies only on element-wise products, simplifying the model and enabling parallel computation. - Channel Mixing: Adapts the Gated Linear Unit (GLU) to use BitLinear layers with ternary weights, removing expensive MatMuls while still mixing information effectively. - Hardware Efficiency: Introduces a fused BitLinear layer that optimizes RMSNorm and BitLinear operations, improving training speed by 25.6% and reducing memory usage by 61%. - Experimental Results: Shows the MatMul-free LLM performs competitively with current models on various tasks, with potential to outperform them as it scales. - Custom FPGA Accelerator: Built to leverage the lightweight operations of the MatMul-free LLM, processing billion-parameter models efficiently at 13W—demonstrating human brain-level efficiency and showcasing the potential for brain-like efficiency in future lightweight LLMs. Exciting time to be alive! #5iR KamiwazaAI

14

4 Comments

OpenAI released the price of GPT-4o mini which will enable a lot of business to use the LLM API more but it is still not cheap for big use cases. Here's a brief overview of the pricing for small LLM models from the main suppliers: 1. Claude Haiku Input : $0.25 / 1M token Output : $1.25 / 1M token 2. Llama-3 8b (Groq) Input : $0.05 / 1M token Output : $0.08 / 1M token 3. Gemini 1.5 Flash Input : $0.35 / 1M token Output : $1.05/ 1M token 4. GPT-4o mini Input : $0.15 / 1M token Output : $0.60 / 1M token For comparison, the standard GPT-4o pricing is : Input : $5.00 / 1M token Output : $15.00 / 1M token The price reduction is significant - but - keep in mind that using a Small Language Model impacts heavily output quality. My main concern with this price is knowing how computationally expensive llms are and the green factor for such big models are not that great. How are they making it available so cheap?

6

Can we trust machines to judge machines when it comes to LLMs? There's an ongoing battle between manual and automated evaluation of LLMs. Manual evaluation is predominantly used by many companies, especially large ones with sufficient budgets. This approach involves internal QA teams or external consultants who manually create test cases and evaluate the outputs in spreadsheets. This process is time-consuming, but can provide insights into: - model's performance - specific failure modes such as hallucinations - tone discrepancies - biases - copyright violations - privacy issues Smaller companies without such budgets often resort to using expensive engineering time to manually review outputs, which can be inefficient. The primary advantage of manual evaluation is the accuracy and reliability of human judgment. But it's not scalable and introduces human biases, which can affect the consistency of the evaluations. Also manual processes cannot keep up with the rapid development cycles and the volume of outputs generated by modern LLMs. Automated evaluation aims to reduce the reliance on human labor by employing metrics and models to assess LLM performance. Techniques used in automated evaluation include: - Metrics-Based Evaluation: It utilizes established metrics such as BLEU score, ROUGE score, and perplexity. These metrics help assess aspects like fluency, accuracy, and relevance. But these metrics often do not correlate well with human judgments and can be inadequate for comprehensive evaluation. - Heuristic-Based Functions: These are specific functions to test particular capabilities or behaviors of the LLM e.g. toxicity detection or PII (Personally Identifiable Information) detection. These approaches can be tailored to specific use cases but may lack generalizability. - Model-Based Evaluation: It uses another trusted language model to evaluate the outputs of the target LLM. This can automate the evaluation process but requires careful calibration to ensure reliability. Automating this process can be a huge unlock. But developers attempting to create their own automated evaluation systems report very low confidence in the results. Lot of opportunity on this front.

3

1 Comment

The Voice AI space keeps getting hotter. (In case you missed it) ➜ Kyutai launches GPT-4o's alternative 𝑀𝑜𝑠ℎ𝑖 Similar to GPT-4o, 𝑀𝑜𝑠ℎ𝑖 is a multiamodal LLM. Capable to listen, hear, and speak. Expect similar demos from other teams soon. ➜ Daily launches Pipecat Pipecat is a framework for conversational AI. And it is open-source. I tried the demo, and it was super exciting. I'm hoping to see more of this soon. See a demo here: https://dub.sh/TbPwadu ➜ Vapi launches SIP Telephony Support We can now have Voice AI in legacy systems. For example, a PBX or VoIP provider. See the update: https://dub.sh/FmS2PUF It's super exciting to see how fast things are moving. ~~~ ♻ Re-share for your friends in Voice AI

53

2 Comments

One line of software can impact a billion rows of data. A short SQL statement can delete a table, reformat date into the US format, or compute the average quota attainment by account executive by region over a company’s history. As data has become a critical component of analytics & production systems, data engineers require more sophisticated tools to manage their data transformations. Tobiko Data is that solution. Tobiko enables modern data transformations by calculating only what’s needed. At small scale this may not seem important. But for Harness, this innovation saved 30-40% of their cloud data warehouse spend & reduced model build time by 80%. DreamHaven, a gaming company, found Databricks’s Delta Live Tables 3.5x more expensive than Tobiko to run the same pipelines. Existing systems recalculate too much because they don’t understand data lineage & the relationships between tables. So they recalculate too much data to ensure correctness, substantially increasing data warehouse spend. More than that, Tobiko enables data engineers to operate with the same discipline as software developers: creating development environments, writing tests to ensure code performs as intended, & simulating changes to ensure one line of code doesn’t wipe a billion rows. Initially, Pipe planned to have multiple Databricks environments : dev, prod, and staging. They wanted to segregate the data; however, with Tobiko’s architecture, they only needed one environment. This simplified their architecture & reduced cost. Migration is straightforward : Tobiko’s platform is DBT compatible out-of-the-box. Tobiko is founded by a team from Airbnb, Netflix, Google, & Apple who have faced the challenges of scaling data infrastructure & developed tools to solve the problems they faced. We’re excited to partner with Toby, Tyson, & Iaroslav by leading their Series A financing, alongside Unusual Ventures, Jordan Tigani from MotherDuck & George Fraser from FiveTran. If you’re interested to learn more about Tobiko & the modern way to shape data, talk to the team here : https://lnkd.in/gBFRh-4q

73

6 Comments

Hard-coded model routing takes away intelligence from your system instead of adding to it! Interesting read below on how Ema Unlimited's Fusion model is a potent solution for the heterogeneous multi-model future of AI in the enterprise.

14

1 Comment

#pgvector is an open-source #PostgreSQL extension that enables efficient handling of vector data, making it ideal for machine learning and applications requiring vector-based computations. #PostgreSQLPerformance @MinervaDB #dba https://zurl.co/EnCm

2