Hop Labs

PSA: We're hiring ML Engineers and ML Researchers at Hop (apply here: https://lnkd.in/enXyNcfZ), but do not work with external recruiters. If you are getting a message offering you a job or an interview at Hop from someone that does not have a verified @hoplabs.com email address, please report it as a scam. We've gotten a handful of candidates forwarding us emails from folks we've never heard of claiming that they represent us and using vaguely similar email addresses.

ONLINE CONNECTIONS ARE A REMOTE SUBSTITUTE FOR REAL LIFE Know anyone who spends a lot of time in front of a screen? Our whole team does. Since you’re reading this online, there’s a good chance you do too. How much effort do you put toward balancing that with in-person social time? Technology offers so many benefits in our day-to-day living, including social connection opportunities that wouldn’t be possible without it. However, awareness of its limitations and downsides, as well as our fundamental need for real-life connection, is key to staying healthy. https://lnkd.in/eknW7bvw Subscribe to our list for biweekly AI insights to your inbox: https://lnkd.in/g6JDWQXa #remotework #onlineconnection #mentalhealth #lonelinessepidemic #isolation #socialmedia #cvpr

RAG (Retrieval-Augmented Generation) applications, while powerful, can face significant challenges when the retrieved information is irrelevant or inaccurate. Here are some key pitfalls: - Irrelevant retrievals can lead the model to generate incorrect or tangential information, reducing the overall quality and usefulness of the output. - When faced with unhelpful context, models may be more likely to fabricate information to bridge the gaps, potentially presenting false information as fact. - Models may present incorrect or irrelevant information with high confidence if it's based on authoritatively written but off-topic retrieved content. - Processing and attempting to incorporate irrelevant retrievals wastes computational resources without improving output quality, potentially slowing down the system.

Many promising AI projects die in the "valley of death" between POC and production. To get past this, reframe your POC phase. Instead of demonstrating cool capabilities, focus on rigorously defining success criteria, building evaluation infrastructure, and identifying failure modes. A successful AI POC isn't a flashy demo; it's a risk assessment and a clear path to reliable deployment.

THE TOOLKIT FOR GETTING LLMS TO DO WHAT YOU WANT, PART 2 When it comes to approaches for guiding the behavior of LLMs in their applications, prompt engineering, fine tuning, and LLM chaining garner the lion’s share of attention in this space, and for good reason – they don’t require extremely deep technical expertise, and they support fast iteration cycles. However, they don’t encompass the full scope of techniques that can be or will be brought to bear in the creation of LLM applications in the coming years. In this post, we cover three more tools, from de rigueur for complex LLM applications to speculative techniques that may not be production-ready for some time yet. https://lnkd.in/eXDNsZsZ Subscribe to our list for biweekly AI insights to your inbox: https://lnkd.in/g6JDWQXa #PromptEngineering #MachineLearning #LLM #ArtificialIntelligence #AITechniques #FineTuning #LLMApplications #AIInnovation #TechInsights #AIDevelopment #FutureOfAI

Federated learning is a method of training AI models that keeps data where it originates, rather than gathering it all in one place. It works by having multiple participants - like organizations or devices - collaborate on improving a shared model without exchanging their raw data. Instead, each participant uses its own data to calculate improvements to the model, then sends only these improvements to a central system. This central system combines all the improvements to update the shared model. This approach solves several important problems in AI development. It protects privacy by letting sensitive information stay put, and it allows the use of diverse datasets that can't be centralized due to legal or practical reasons.

Three characteristics of successful GenAI use cases: - Availability of data at onset and/or through data-generating processes - Situations where even imperfect solutions are valuable - Well-defined, quantifiable objectives for performance evaluation

If you're an investor in the healthcare space with questions about AI, we have answers for you! Join Hop's founder Ankur Kalra and Catherine Davies from Monticle Ltd tomorrow, July 2nd, at 8am EDT/1pm BST -- they'll be addressing some of the most common questions they've heard from investors like you. We're aiming to limit this conversation to folks who are active in this space -- if this is you, please register to join us: https://lnkd.in/eueQgi8d

At Hop, we love holidays based on math. Happy Tau Day, everyone! Wishing you double pi(e) today!

Developing production-ready LLM applications is like an iceberg - what's visible above the surface is just a fraction of the total effort required. Above the Surface - Model integration - Prompt selection - UI/UX Below the Surface - Robust error handling and fallback mechanisms - Prompt repository and testing - Data privacy and security - Scalable infrastructure - Monitoring and observability - Continuous model evaluation and retraining pipelines - Guardrails and safeguards - User feedback loops and iteration processes