Alex O.

The prevailing methods to make large language models more powerful and amenable have been based on continuous scaling up (that is, increasing their size, data volume and computational resources) and bespoke shaping up (including post-filtering, fine tuning or use of human feedback). However, larger and more instructable large language models may have become less reliable. By studying the relationship between difficulty concordance, task avoidance and prompting stability of several language model families, here we show that easy instances for human participants are also easy for the models, but scaled-up, shaped-up models do NOT secure areas of low difficulty in which either the model does not err or human supervision can spot the errors. #genAi #LLM #LLMs

9

1 Comment

Interesting discussion and episodes on the use of the AutoGen framework for Agentic Workflows at the Microsoft Research Forum. The #Agenticframework allows LLMs to allow to perform more complex and intelligent tasks autonomously through one of the key concepts of reflection and autonomy. It has significant impact on practical applications such as personalized medicine, automated travel concierge, autonomous navigation etc.. It was also very interesting to read blogs posted on deeplearning.ai by Andrew Ng. https://lnkd.in/daa28seD

10

2 Comments

🚀 SmolTalk is now available on the Hub! Nothing but the secret recipe of SmolLM2! SmolTalk is a mix of different public and synthetic datasets, including Magpie Ultra using distilabel (for those curious about the outcomes of synthetic data 🧠). It improves the instruction following and reasoning and also includes rewriting, summarization, and function calling. Explore the dataset here: https://lnkd.in/duwVM-4q Haven’t tried SmolLM2 yet? Don’t just take our word for it—experience its performance firsthand: https://lnkd.in/dtym9RWS

30

#Opera Browser Enhances Multimedia Controls and Introduces #AIPoweredImageGeneration #Opera announced today the release of #OperaOne’s second version in #developerbeta, featuring new multimedia controls, split tabs, and enhanced #AI capabilities. This update introduces new themes and design elements for a refreshed look. Previously, multimedia controls were on the sidebar, but the new version offers a resizable, floating window for these controls, matching the browser's theme. The multimedia player, depending on its size, will have play/pause, next, previous, pop-out, and volume options. Additionally, #OperaOneR2 introduces split tab windows, allowing users to work on two web pages simultaneously—a feature also seen in newer browsers like #Arc and #SigmaOS. A redesigned indicator in the form of an underscore highlights recently closed tabs, with light or dark traces based on how long ago the tab was closed, appearing for users with more than 30 tabs open. Opera has been advancing its AI capabilities, and Opera One R2 integrates these features more deeply. The company has experimented with its "#Aria" assistant, #AIpowered summarization, and the use of #largelanguagemodels (LLMs) locally. Previous experimental AI features like image and voice generation and image understanding are now included in Opera One R2. The new version also introduces a page context mode, allowing users to ask Aria questions about a webpage, such as finding, translating, or summarizing specific information. Opera plans to release Opera One R2 to a broader audience later this year. #Opera #Provelopers #OperaOne #DeveloperBeta #MultimediaControls #SplitTabs #AI #Aria #BrowserUpdate

3

New Blog Series Alert: Responsible AI from Principles to Practice! 🌐by Carlotta Castelluccio April Speight Are you a developer eager to dive into the world of Responsible AI (RAI)? Teh team has just launched an exciting blog series that covers everything you need to know about RAI, from conceptual overviews to practical examples within the Azure ecosystem. 🔍 What to Expect: Ethical AI Development: Learn how to build AI systems that are ethical, secure, and inclusive. RAI Toolchain: Discover the tools available in Azure to help you innovate responsibly. Practical Insights: Get hands-on examples and strategies to mitigate risks and safeguard your AI systems. Don’t miss out on this opportunity to enhance your skills and contribute to a more responsible AI future. Check out the first post now and stay tuned for bi-weekly updates! 👉 Read the blog series https://lnkd.in/eM-uvvxF #ResponsibleAI #AI #RAI #MachineLearning #Azure #EthicalAI #TechCommunity

4

Had a great start of the week in Berlin at #OpenSearchCon 2024. The videos, including those of the Zeta Alpha's talks on LLM Evaluation with RAGElo and on fine-tuning embeddings for vector search are online now. Fine-tuning embeddings 4 vector search (Jakub Zavrel): Using open source embeddings like e5 and fine-tuning them on in domain synthetic data yields significant improvements over out of the box one-size-fits all embeddings from OpenAI or other vendors. https://lnkd.in/gBT7UHMH RAGElo: An Elo Rating-based Evaluation Toolkit for RAG (Jakub Zavrel & Fernando Rejon Barrera) Using LLMs as a judge is the key to driving quality of RAG applications forward while keeping your domain experts happy. https://lnkd.in/gPMjzTup Challenges in moving Gen AI apps to production (Fernando Rejon Barrera) Interactive session on the multifaceted mine-field of developing and deploying generative AI applications in a way that is both technically sound and aligned with business and safety requirements. https://lnkd.in/gfw4b9TD

62

2 Comments

The matter of abstractions The realms of knowledge are not flat, but the sky is not the limit, hence the need of mapping abstraction scales. https://lnkd.in/ePbAGUcf

13

"There’s a growing trend of “open washing” that is undermining the meaning of open source". This is a quote from our 2024 Hype Cycle for Open Source: https://lnkd.in/ePqPhRm8 published by Oleksandr Matvitskyy, Anne Thomas, Nitish Tyagi, and Mark Driver Debate about the "open washing" trend continues, most recently from Steven Vaughan-Nichols in The Register: https://lnkd.in/epMYF2Xk End-user organizations are caught in the middle. Earlier this month, Nitish Tyagi and Chris Saunderson published this advice: "Impact & Risks of Open-Source to Source-Available Licensing Shifts". This is a topic that will only increase in importance in 2025 https://lnkd.in/e6pJFnyc [Gartner sub req'd]

33

3 Comments

hype-o-ware (n): A sub-genre of vaporware, distinguished by conformance with current tech hype cycles. e.g., "We'll solve human aging via crypto" or "AI mumble mumble".

73

6 Comments

Really excited by the release of GPT-4o mini. As well as being 60% cheaper than GPT-3.5 and way more performant, the longer output token limit will come in handy for our podcast generation tool. Perhaps most interesting will be the opportunity to fine tune; fine-tuned open weight models have been shown to outperform GPT-4 for specific use cases, so I am really looking forward to seeing what fine-tuned GPT-4o mini can do! AI generated podcasts: https://lnkd.in/ev_ceij6 Finetuned models beat OpenAI’s GPT-4: https://lnkd.in/eFccUKir

8

Domain Events published publicly should have a full namespace, even if it's minimal. For example, the internal model is: co.nucoverage.rate.model.calculator.RateCalculated Company is NuCoverage The context is Rate Context The public event name is: nucoverage.rate.RateCalculated These examples are from my book "Strategic Monoliths and Microservices: Driving Change Using Purposeful Architecture."

108

26 Comments

The new episode of Spilling The Source is out. this time I talk to Marco Vinciguerra about Scrapegraph AI. A site scraper that uses LLM models for processing. https://lnkd.in/eJ2m6APz

5

2 Comments

Google DeepMind unveils technical architecture behind NotebookLM and Illuminate's multi-speaker podcast audio generation: Then we finetuned it on a much smaller dataset of dialogue with high acoustic quality and precise speaker annotations, consisting of unscripted conversations from a number of voice actors and realistic disfluencies — the “umm”s and “aah”s of real conversation. This step taught the model how to reliably switch between speakers during a generated dialogue and to output only studio quality audio with realistic pauses, tone and timing. https://lnkd.in/dNjJRG7Q #texttospeech

5

Today, #TheSequence dives into AI21 Labs Jamba: Not Just Transformers: Jamba is New LLM that Brings the Best of SSMs, Transformers, and MoEs in a Single Architecture https://lnkd.in/eHwQ4RUZ #generativeAI #artificialIntelligence #LLMs

8

New video! Are You Choosing the WRONG LLM for Your Project? Learn more 👇 Video 2/2 with Integrail :) #llm #llms

13

I just completed the GCP carbon aware computing class, which was very insightful. The usage of electricity maps was eye-opening. Try out the coordinates for the US and compare them with Germany. You can find the high amount of coal and wind usage in the German electricity grid and the transfer back and forth within the European countries participating in the grid. You can also see that, although Germany does not produce any electricity using nuclear, there is still a small amount of electricity based on nuclear plants in other countries transferred into the grid. #GCP #CarbonAwareComputing #ElectricityMaps 🌍⚡️

19

3 Comments