AQ22

We are ready for Silicon Valley, but is it ready for us? 😄🚀🇺🇸 What has happened during the last half a year at AQ22: - We pivoted and then niched down to finance. - Found amazing mentors, who come in with strong industry expertise. - Continued building our Financial LLM. - Had over 40 meetings with prospects for pilots. - Talked with the world's biggest banks and also SME banks. - Signed 3 NDAs and have 2 pilots ready to start. - Finalized all the legal aspects of the business for future investors. - Finished our Plug and Play Europe program. - Were selected to be showcased in World Summit AI (happening in 🇳🇱). - Were selected to participate in the Plug and Play GOAL (Global Overseas Acceleration & Learning) this autumn. But we also had many down moments too, many rejections, many follow ups which never got an answer, and other fails, which we count as lessons on the way to the top. It was a tough, tough road, and when I put it here, it looks like it all simply happened one after another, but there were many roadblocks 🚧 along the way. I remember one week when we were totally confused about whether there was a market fit. 😄 But hey, I guess everyone goes through such moments in the beginning. I keep a notion file with all the challenges and wins so I hope to share it one day with all of you. For now, I am so so so excited to be going to the USA and accelerating our growth even more. Hope you are ready to follow our journey here, on LinkedIn. 🚀 Thanks for all the support.

Our CTO, David Mataciunas, has been featured in The New York Times, along with other researchers, in a new study from the Data Provenance Initiative. Their research uncovers a critical shift in AI development: the drying up of accessible web data for training AI models. 🌐📉 Over the past year, many key web sources have restricted the use of their data, sparking what we term an "emerging crisis in consent." Their study examined 14,000 web domains and found that 5% of all data, and a staggering 25% of high-quality data, has been restricted. 🚫💾 This tightening access impacts not only AI companies but also researchers, academics, and non-commercial entities relying on public datasets. As Shayne Longpre, the article's lead author, states, "We're seeing a rapid decline in consent to use data across the web." This evolving landscape signals significant challenges and calls for innovative solutions in how we collect and use data for AI training. Incredible team: Shayne Longpre, Robert Mahari, Ariel N. Lee, Campbell Lund, Naana Obeng-Marnu, William Brannon, Nikhil Singh, Manuel Cherep, Joanna Materzynska, Sara Hooker, Sandy Pentland, Jad Kabbara, Caiming Xiong, Hanlin Li, Stella Biderman, and many more contributors. Article by Kevin Roose 🚀 The article is in the comments ⬇️ #ai #largelanguagemodels #machinelearning #deeplearning

WANTED: Business Analysis Intern Join us as an intern and dive into the exciting world of market analysis (or at least the startup ride 🦄)! Help us uncover trends, evaluate financial solutions, and analyze competitors in the financial sector. THE JOB: - Market analysis of financial institution positions - Old financial institutions’ solution providers - Market analysis - Competitor analysis and other fun stuff if you like financial markets. HOMEBASE: Lithuania Right now, the internship is not paid but has the potential of a bonus once the autumn round closes. If you want to work alongside David Mataciunas, send your message to him here on LinkedIn or email to [email protected] / [email protected] Please, help us find a diamond in the rough. 💎 Cc: Tech Philomaths, Plug and Play Europe, ISM University of Management and Economics, Eigirdas Žemaitis, PhD. #intern

I am proud to have participated in the Data Provenance Initiative's study, "Consent in Crisis: The Rapid Decline of the AI Data Commons." Today, we are releasing a second large-scale audit of 1800 LLM training datasets and the 14k of the websites. Key insights: 🔒 Rising Restrictions: Restrictions on AI's use of web content have surged, with a 500% increase from April 2023 to April 2024. Nearly half of the C4 dataset is now subject to Terms of Service, forecasting further increases by 2025. 📉 Declining Data Commons: The availability of high-quality pretraining sources is rapidly diminishing. By April 2025, an additional 6-10% of the highest-quality tokens in AI datasets will be restricted. ⚖️ Varied Impact: Different AI companies face varying levels of restrictions, with OpenAI's web crawler being the most restricted (91%) and Meta's the least (52%). 🚧 Protocol Challenges: Current web protocols, such as robots.txt, are ineffective in accurately conveying consent intentions, leading to inconsistencies and potential misuse. This study underscores the urgent need for better mechanisms to communicate and respect data consent, impacting AI development and ethical considerations. A big thank you to our incredible team: Shayne Longpre, Robert Mahari, Ariel Lee, Campbell Lund, Naana Obeng-Marnu, William Brannon, Nikhil Singh, Manuel Cherep, Joanna Materzynska, Kushagra T., Sara Hooker, Sandy Pentland, Jad Kabbara, Caiming Xiong, Daphne Ippolito, Hanlin Li, Stella Biderman and many more contributors!!! #ai #ethics #datasets #machinelearning

The financial sector is on the verge of significant change! LLM integration across all functional areas is expected within five years. KPMG reinforces this opinion, stating that 60% of financial executives expect their first GenAI solution to be implemented in 1-2 years. The experts at McKinsey & Company also predict GenAI's important impact on the banking industry: an additional $200 to $340 billion per year in revenue, mainly due to improved productivity. This is exactly why we are building our financial LLM, which will serve financial institutions on a daily basis. The big opportunity lies ahead of us. P.S. We are raising our first round in autumn. If you want to keep updated, ping Kotryna Kurt or David Mataciunas. #ai #financial #finance #banking Source: Dynamiq / Vitalii D.

Baltic Founders Mingle S01E01 by Tesonet. 50 leading Founders from Latvia, Estonia & Lithuania under one roof at CYBER CITY for the first time ever. It was great to hear Tomas Okmanas, Taavet Hinrikus, Rytis Vitkauskas, and James Janis Berdigans discussing on bridging the gap between the Baltic countries. Our CTO, David Mataciunas, got a chance to chat about AQ22 and AI opportunities which will bring more competitive advantage to the region. #event #ai #founders

Why is the function calling so important for AI agents? 🤖 Function-calling agents can automate complex tasks that involve interacting with multiple digital services and applications. However, they require high data quality, data diversity, and collection scalability. Salesforce presents an automated data generation pipeline to synthesize high-quality datasets for function-calling applications and outperforms GPT-4 models and other state-of-the-art models on the Berkeley Function-Calling Benchmark. This is a game changer because most of the function-calling datasets for fine-tuning are not diverse. In AQ22, we are building AI agents for the financial industry, and function calling is super important to collect information that is important for specific tasks. We are definitely releasing some function-calling financial models for everyone soon! 🚀 #ai #salesforce #largelanguagemodels #deeplearning

Product-Market fit - something you can read about in countless textbooks, articles or blogs. However, I like the framework made by Sequoia Capital the most. HAIR ON FIRE HARD FACT FUTURE VISION You can choose one of these 3 paths. For AQ22, we are going with the Hard Fact road - people might be used to working with what they have because there is nothing better. YET. The pain is there, however, sometimes it is not worth solving with new tools, so hiring more people will do. We are here to change that. Anyhow, if you are a startup or planning to build one, take a good look at this framework to see where you will be placed. #AI #productfit #marketfit #startups #enterpreneurship

GRADUATED! 🚀 We are happy to be one of the Plug and Play Tech Center graduated startups and continue our journey. It has been 3 months with workshops, mentors, startup gatherings, pitching, and much more! The team is very grateful to Raimondo, Anastasija, Andrej, and Leonardo for all the support and the right push. Also, thanks to Tom and Seymur for mentoring us along the way. 🚀 Startups you should be watching in the next months: - AQ22 (us, duh 😎) - Biohifas - Boop - micapass - Kloosive - Uncovered — Competitive Intelligence - WhiteBridge.ai - InGenes - use your genes 🧬 - SpotGo - eTOLLs EU The road was like a rollercoaster - ups and downs, but eventually, that's what every startup journey looks like! Stay tuned. #aq22 #plugandplay #accelerator

Are you in the money game? 💰 There are a few ways how your job could soon change. One survey found that 75 percent of junior financial services workers believe AI could automate up to a quarter of their workload. How Will Generative AI Change Finance Work? 👇 Generative AI tools will streamline transaction processing by automating many of the routine tasks, such as data entry, validation, and reconciliation. This will mean banks and other financial institutions can offer faster, more reliable, and perhaps even cheaper services to their clients. - Detecting and preventing fraud - GenAI can do this far more quickly than any human and can also report results in a way that makes it simple for a human risk assessor to step in and understand why a particular transaction could be suspicious. - Managing investments - AI algorithms can assist with analyzing markets, assessing the performance of companies and other assets, identifying trends, and evaluating the effectiveness of investment strategies. - Insurance underwriting - GenAI is able to understand and interpret information around risk, providing more detailed and accurate assessments that will help underwriters make more informed decisions. - Regulation and compliance - GenAI will help with analyzing and interpreting complex regulatory texts and legal documents. - Customer services - chatbots and automated customer service systems have been frustrating in the past. But a new generation of ChatGPT-like chatbots, powered by large language models (LLMs), will answer many questions more quickly and efficiently than human operators, improving customer satisfaction and retention rates. So much to do and so many opportunities for smart integrations. We are rolling up our sleeves at AQ22. Source: Forbes / Bernard Marr #AQ22 #AIagents #finance