Suggestions
Anita Kirkovska
Growth at Vellum.AI - Running ThePrompt
Anita Kirkovska is the CEO/CMO at The Shelf, a no-code and AI agency specializing in developing MVPs within 4 weeks. With a track record of working with 15+ clients, Anita has also created innovative products like VoicePen.AI, which converts audio and video content into blog posts.
Anita holds a master's degree in Computer Science and is a Fulbright scholar. She is passionate about utilizing her expertise in growth, project management, AI, and no-code to address real-world challenges. Constantly exploring new technologies, Anita collaborates with fellow creators to achieve collective goals and seeks fresh approaches to support teams in reaching their objectives.
Her educational background includes studying Data Science Datanaut at NASA Datanauts, International School for Young Leaders, a Graduate program at the University of Central Florida, and a Bachelor's degree from the Faculty of Computer Science and Engineering in Skopje.
Anita's professional experience spans roles such as Maker; Growth at VoicePen AI, Founding member; PM, Growth at Intersect Labs, Growth Product Manager at Ampstor, Fulbright Scholar, Growth Hacker at Adeva IT, Marketing Assistant at NewMan's Business Accelerator, and Project Management Lead at NASA International Space Apps Challenge.
Highlights
When LLMs actually do what they should in production
OpenAI found hidden features in GPT-4's mind—16 million of them.
When we talk about "features" in the context of LLMs, we mean the distinct patterns or attributes that the model recognizes and uses to process information.
These features help the model understand and generate language. For example, a feature could be a specific grammatical structure, a common phrase, or a particular type of information like dates or names. By identifying these 16 million features, OpenAI has made it easier for humans to understand how GPT-4 thinks and makes decisions.
For example, feature 63541 focuses on human flaws and activates sentences like this:
"Most people have wonderful days, glimpses of what we perceive to be perfection, but we can also have truly terrible ones. You're not alone."
Another feature, feature 9260, addresses price changes and activates sentences like this:
"Prices rose $0.16 after Federal Reserve Chairman Ben Bernanke suggested that the U.S. central bank would continue its policy of low interest rates to help spur job creation and economic growth, according to an Associated Press report."
Although LLMs are already something we can't live without, we still don't know how they actually work.
The "Golden Gate" experiment by Anthropic that I posted about yesterday and OpenAI's research help us understand neural networks and guide language model behaviors more responsibly.
You can find more feature on this link: https://lnkd.in/gipQecgA
Sign up for our newsletter to learn about LLMs and the latest research: https://lnkd.in/d2beZB_U

