Recently attended an AI Dinner
hosted by Snorkel AI
with their Co-Founder and Head of Technology, Braden Hancock.
Great event at Portale Restaurant in NY
Snorkel AI is a technology startup empowering data scientists and developers to quickly turn data into accurate and adaptable AI applications powered by programmatic labeling and annotation.
It seemed like there was representation from every financial firm and interest expanding to the life sciences for the usage of AI.
- AI Models - May eventually become commoditized where looking at the end to process, where there is a sweet spot is the tagging, annotating and categorization of data to create various test datasets to train multimodal ai
- Good Data is Everything - Imagine if all of the AI models are commoditized the differentiator is the data you own. If you wanted to build the most best AI around physics, imagine if you could scan in the mind of Einstein as your reference data. If you want to build the best model on exercise, imagine if you had the experiences all mapped of the worlds greatest Olympians. The models are how we interpret, organize and use the data to provide compelling responses, but the content is largely a function of the data you ingest and the precision it provides. This concept is further explored in the recent Netflix series JUNG_E
about CombatAI and how we mine the right data to drive good AI results.
- Continual Feedback Loops is Key - To continually learn and evolved, there must be a way to measure the confidence score of an AI model. Simplify mechanisms to determine if the result was good or bad enables the models to correlate what is common and self-health to isolate bad logic or outlier data and then adjust the models.
- GANS Theory (Generative Adversarial Networks) - The tech continues to expand with what used to be neutral networks to more and more companies having adversarial networks. What this means in simple is that you have 1 AI model to try to attempt to do something like create code, make artwork, etc and then you have another AI model trying to validate it. Find gaps in the code, vulnerabilities or detect if the artwork is fraudulent. It seems to be easier to have 2 models go against each other then to continually try to fix 1 model in isolation.
- New Paradigm Shifts in Staffing - As AI grows to to take over the bottom workforce, executive leaders need to think actively how they re-purpose staff to other value add areas. This is critical because while a less progressive leader may reduce headcount, the challenge in the long run is that this reduces career opportunities to more senior roles resulting in a gap in talent pipelines to both mid-level and senior level roles
- Multi-LLM is just as important as Multi-Cloud - As new models and features in LLM are created, it is critical to not put all of your eggs in one basket. ChatGPT, Google Gemini and others are evolving how they manage their models, the features they provide and the underlining data sets you can leverage as part of the overall consumption of the service.
- Digital Exhaust - In the digital landscape it is our data exhaust that is driving the monetization of big data. Big companies like Google, Amazon and others have been giving free services for a while now with the trust-model that they get access to our digital lives and footprints which empowers more successful up-selling where you advertise the products people need based on the data around them
- Cell Phone Data - It shows people congregating in the same location, it shows the locations you go and this ultimately allows us to infer likes and dis-likes. If you goto Chinese school or yoga class, the geolocation shows this and it indicates the types of things you like.
- Internet of Things - In the future our things will communicate and share data more freely. Your fridge will know if your milk is going to expire, but indirectly based on what is in your fridge it can guess your diet and overall health. Your bed can tell if you slept in it or not and if you are tossing and turning. This could then indicate certain medication or other services that may help you get a better night of sleepThere is a great booking that has provided a number of examples on this including the ones above by Bruce Schneier called "Click here to kill everybody
" where Trust Models and how we think about trust is changing with more people comfortable with corporations owning our data and diluting the exiting concepts of privacy for the sake of new services, innovation and efficiencies. It is not wrong or right, just different and part of the times.
Cyber Security, Technology and Risk Executive | Board Member | CISSP | Speaker | Patent Holder | Product Manager
9 个月Thanks Jeanette. Great event
Field Marketing Leader ? AI ? Start-ups. Ex-Snorkel AI, Affirm, HP, Intuit
9 个月So great to have you join us at dinner Miles Dolphin.