LLM Hallucinations ??

LLM hallucinations have forever been a thorn in the side of researchers and developers, remaining an unsolved problem. However, there’s hope now with a lot of?new techniques emerging in the space, claiming to address this challenge to an extent.?OpenAI?function calling, Microsoft?Guidance, and?GrantSlatton’s patch?for?llama.cpp, alongside?SudoLang,?prlang?and?Jargon?are some of the methods that enforce a context-free grammar or a?JSON schema.

Out of all these techniques,?LMQL?by?SRI Lab?seems to be?gaining the spotlight.?

Developed by?Luca Beurer-Kellner,?Marc Fischer, and?Martin Vechev, this new technique offers a concise and intuitive syntax. It also helped them reduce the compute costs by up to 80%, as mentioned in their research paper ‘Prompting is Programming: A Query Language for Large Language Models’.?

“LLMs+PLs is a very interesting field right now, with lots of directions to explore,”?said?Beurer-Kellner. This offers users the ability to express both common and advanced prompting techniques in a simple and concise manner.?

But, why use LMQL, when you can directly code your queries in Python or C++ or any other programming language??

With LMQL, developers and researchers can avoid writing complex code for text concatenation and output parsing, as these processes are streamlined within the tool. As a result, you can focus more on the core logic of your project and spend less time dealing with cumbersome implementation details.?

Read the full story?here.

Kicking OpenAI Out

German computer scientist?Josef “Sepp” Hochreiter?has an interesting personality. He believes that his current research work, which he calls ‘XLSTM’, is so much better than OpenAI’s GPT models that it would kick them out of the LLM supermarket for good.

In an exclusive interview with?AIM, he recounts the struggle of his?research paper?being rejected by NeurIPS a long time ago, and how the very same research is now revolutionising the deep learning and AI landscape.?

Talking about his latest research work, where he alongside his team is feeding every transformer right now on smaller datasets combined with LSTMs, he said: “We are so much better than GPT (generative pre-trained transformer) and want to kick OpenAI from the supermarket in autoregressive language modelling.”?

Read the complete story?here.?

India ?? ChatGPT?

India has been one of the key markets driving ChatGPT’s early success, with even chief?Sam Altman acknowledging it?during his recent visit to the country. But now with the release of the ChatGPT Android app, the company is going to experience an even greater outpouring of affection and support.

Within 24 hours of its launch, the app has garnered 1 million downloads. According to?Statista, Android held a share of 95.26% of the mobile operating system market in India, followed by Apple’s iOS – a distant second, with a 3.92% market share in 2022. The prospect of reaching and serving such a massive user base in India?holds great promise for the future of ChatGPT.?

Hype, Exit, Repeat

If you look at the generative AI investment landscape, one name seems to be everywhere – Andreessen Horowitz aka a16z – a stage-agnostic venture capital firm that backs entrepreneurs building the future through tech.?

Interestingly, the firm is also well-known for leveraging effective PR to create hype around tech. It has done so for blockchain, crypto and metaverse in the past. Other examples include Coinbase, Airbnb, Affirm, Instacart, Netscape, and Skype — leading to high valuations.?

After the exit, however, the valuations of these companies often decline. If the barrage of their well-crafted and researched blogs around generative AI is anything to go by, Andreessen Horowitz is looking to do the same once again.?Read on.