Andreas Stöckl

Featured in: Favicon

medium.com Favicon

towardsdatascience.com Favicon

datadriveninvestor.com

Articles

New Functions for Mistral AI

Nov 24, 2024 | medium.datadriveninvestor.com | Andreas Stöckl |Import Os

AI Alternative from EuropeThe European ChatGPT alternative ‘Le Chat’ from Mistral (https://chat.mistral.ai/) has many new features. The French-based European AI hopeful is developing various new functions and options that make the language model and chat a natural alternative to ChatGPT. All new functions in the chat are freely available. The Mistral product outperforms the free versions of all competitors, such as ‘ChatGPT’ from OpenAI, Claude from Anthropic, and Gemini from Google.
Unveiling the Arithmetic Capabilities in Large Language Models

Sep 21, 2024 | medium.datadriveninvestor.com | Andreas Stöckl

Explaining LLMsLarge language models (LLMs) have recently demonstrated capabilities far beyond their initial purpose of generating text. From solving mathematical problems to writing code, their range of competencies continues to expand. However, the inner workings of these models, specifically how and where they store such specialized abilities, remain an enigma.
Watch out for satire! - Caution AI!

Sep 11, 2024 | medium.datadriveninvestor.com | Andreas Stöckl

How suitable are generative AI systems for satire? Satire is a particular challenge for AI systems. To test the possibilities of current AI systems, I have started a project that generates and publishes AI-generated satirical news. Under the website:Satirical news items generated completely automatically by AI are published daily. Similar to satirical news portals such as https://dietagespresse.com/ or https://www.der-postillon.com/, news articles usually related to current events are offered.
New Paper from Google Deepmind

Apr 19, 2024 | medium.datadriveninvestor.com | Andreas Stöckl

Grandmaster-Level Chess Without SearchThe paper “Grandmaster-Level Chess Without Search” explores the possibility of training a chess policy using supervised learning without relying on explicit search algorithms. The researchers aimed to investigate the impact of training at scale for chess and determine if a large-scale transformer model could achieve grandmaster-level chess play.
Comparing Aleph Alpha and Mistral AI

Dec 14, 2023 | medium.datadriveninvestor.com | Andreas Stöckl

A Deep Dive into European AI InnovationThe European AI landscape has been significantly bolstered by two major players: Aleph Alpha and Mistral AI. Both companies have made remarkable strides in advancing generative AI technologies, each with unique strengths and offerings. This article delves into a comparative analysis of their models, focusing on their capabilities, innovations, and impact on AI.
Deploying Large Language Models in Streaming Applications: A Breakthrough in Efficiency and Performance

Oct 6, 2023 | medium.datadriveninvestor.com | Andreas Stöckl

Language models have revolutionized natural language processing tasks, enabling machines to understand and generate human-like text. However, deploying large language models (LLMs) in streaming applications, such as multi-round dialogue, has posed significant challenges. Two major obstacles have hindered LLMs' efficient and stable deployment in streaming scenarios: excessive memory consumption during the decoding stage and the limited ability of existing LLMs to handle longer texts.
The Reversal Curse: A Limitation of Language Models

Sep 25, 2023 | medium.datadriveninvestor.com | Andreas Stöckl

Language models have revolutionized natural language processing tasks, demonstrating impressive capabilities in generating coherent and contextually relevant text. However, a recent research paper titled “The Reversal Curse: Failure of Generalization in Auto-Regressive Large Language Models” sheds light on these models' limitations that hinder their ability to generalize in specific scenarios.
Large Language Models: A New Frontier in Human Mobility Prediction

Aug 30, 2023 | medium.datadriveninvestor.com | Andreas Stöckl

Accurate prediction of human mobility is crucial for various applications, including epidemic modeling, transport planning, and emergency responses. However, due to the scarcity of mobility data and the unpredictable nature of people’s daily activities, achieving precise predictions of their locations remains a challenge. In a recent research paper titled “Where Would I Go Next?
Browsing the Web with Audio Input and Output

Apr 20, 2023 | medium.datadriveninvestor.com | Andreas Stöckl

In this article, I describe a prototype application for a web browser that allows browsing with audio input and then outputs a summary of the content of the pages via audio. This is done using an AI agent built with the Langchain framework. The following video shows an example:Reading BBC News with Audio Browser In contrast to a standard web browser, entering a web address here is not necessary, but only an instruction as to what you want.