VE3’s Post

Learn how Perplexity's AI-powered search engine uses Amazon Bedrock and Anthropic's Claude 3 for accurate, comprehensive answers to user queries. https://lnkd.in/d_n-vyrY

Perplexity Builds Advanced Search Engine Using Anthropic’s Claude 3 in Amazon Bedrock Learn how Perplexity’s AI-powered search engine uses Amazon Bedrock and Anthropic’s Claude 3 for accurate, comprehensive answers to user queries. #GenerativeAI https://lnkd.in/gqnsqmDD

It is getting clearer that LLMs will get super-powerful and soon. The obstacles that we thought will take time are being conquered continuously using methods developed in quasi-opensource. It is a good time to zoom out and think through the next decade or so. Mo Gawdat (ex-Google X) gives a crisp assessment of the changes to come and an empathetic way to adjust to the new realities. It is the best take I have seen this year. https://lnkd.in/g_GyadeG

LLMs and Search Many vertical search engines stand to gain significantly by aggregating diverse information sources, thereby offering customers robust bases for selecting optimal products and making informed decisions to buy them. This process involves gathering data from a variety of sources, including product search indexes, but also reviews, product descriptions, and other structured and unstructured information. See, an example https://lnkd.in/gU_VKX7P, from a toy prototype on how such search can help customers Stanford and Amazon have recently introduced a benchmark to assess the performance of search engines in handling such multifaceted search. https://lnkd.in/dqE24aBe

Quote - "Google, for example, improved the quality of search results long before the advent of large language models, and the Army could apply similar methods to how it handles its large swaths of data, said Riley.  One way the tech giant worked to improve search results was by analyzing which search results were clicked on most often and identifying which results were most useful to the most users. Additionally, the company developed a knowledge graph that “represents widely accepted truths and relationships.” This approach helps ground search results in established knowledge, which subsequently requires less computational power than LLMs." https://lnkd.in/ehbiH_bX

Anthropic’s Claude 3.5 Sonnet model now available in Amazon Bedrock: Even more intelligence than Claude 3 Opus at one-fifth the cost This powerful and cost-effective model outperforms on intelligence benchmarks, with remarkable capabilities in vision, writing, customer support, analytics, and coding to revolutionize your workflows. Read mode on following blog post!

Anthropic’s Claude 3.5 Sonnet model now available in Amazon Bedrock: Even more intelligence than Claude 3 Opus at one-fifth the cost This powerful and cost-effective model outperforms on intelligence benchmarks, with remarkable capabilities in vision, writing, customer support, analytics, and coding to revolutionize your workflows. Read mode on following blog post!

Google's new AI-powered search feature has returned some unreliable, potentially dangerous information. It has suggested that users add glue to pizza and eat at least one small rock a day, and that former US president Andrew Johnson earned university degrees between 1947 and 2012, despite dying in... 1875. So why is it returning wacky answers, and what, if anything, can be done to fix it? Here's what you need to know: https://trib.al/RNkvLHU

Interesting presentation by Jeff Dean from Google about some of the newer trends in machine learning. One big takeaway is how the industry is moving to multi-modal models that move beyond text and are beginning to incorporate audio and video. With mass amounts of good data being a major factor for these models, it seems like Google is in the driver's seat with years of YouTube content to train on. They're also sitting on a financial gold mine should they choose to sell the data for others to train their models on, similar to what we've seen recently from other companies like Reddit, Tumblr, and WordPress. Reddit - https://lnkd.in/gN_aUM7Q Tumblr/WordPress - https://lnkd.in/gXHgH_a7 https://lnkd.in/gxEmj_xV

Google, Microsoft, Mistral - the list of organizations building open-source LLMs is growing fast. The irony, the company with "open" in their name however isn't in that list. Really excited to see LLMs becoming a commodity. Yes, GPT-4 is still the best-in-class, but (a) the alternatives are now already great for a ton of use cases, and (b) give them some more time. The speed at which improvements are made in the open-source field is mindblowing.