Suggestions
Val Orekhov
CTO at Withum Digital
Val Orekhov is the Chief Technology Officer at WithumSmith+Brown, a prominent accounting and consulting firm. He is known for his expertise in various technological domains, particularly in SharePoint and other IT-related areas.1
Val is highly regarded for his intellectual capabilities, with colleagues describing him as one of the smartest people they've worked with.1 His professional journey began around the turn of the century, where he started building websites during the dot-com era.7
In recent years, Val has shown a keen interest in artificial intelligence and its applications in business:
AI Initiatives
- He has been involved in launching a resource focused on safe and efficient AI business solutions.4
- Val has engaged in discussions about Large Language Models (LLMs) and their capabilities, particularly in areas like planning.8
- He has also shared insights on the contrast between engineering and marketing perspectives in the LLM industry.5
Professional Engagements
Val actively participates in industry events and discussions. He has expressed enthusiasm about meeting new people and discussing "boring" AI topics and their applications in employee benefit plans.6
Personal Interests
Outside of his professional life, Val seems to have a passion for cycling. He frequently posts about his cycling adventures under the name "Pedals & Perceptrons," documenting rides along various trails including:
These posts suggest that Val combines his interest in technology (perceptrons being a reference to artificial neural networks) with outdoor activities, showcasing a well-rounded personality that balances professional expertise with personal interests.
Highlights
A small model appreciation post: GLiNER - a model designed for a specific task called Named Entity Recognition (NER). Simply put, NER identifies key pieces of information in a text—such as company names, product titles, or important concepts like regulatory or auditing terms. This task is especially crucial for building knowledge graphs, which help us extract meaningful insights from unstructured business documents.
In many business contexts, accurately identifying these entities is essential. However, most off-the-shelf NER models are tailored for general use cases, like recognizing people or place names, and often fall short when applied to structured business documents, particularly in highly regulated industries. GLiNER stood out by extracting Noun Phrases without any fine-tuning, allowing us to quickly bootstrap our data extraction pipeline.
GLiNER, a modest 300-million-parameter model, was trained on the dataset of a much larger model, Microsoft's Universal NER (https://lnkd.in/ecAKGzzu) , which has 7 billion parameters. Despite its smaller size, GLiNER performs much better with compound, multi-word keywords.
This is a great example of how large, expensive models contribute to the creation of high-quality training datasets, which are then used to train models with more efficient architectures. Kudos to the open-source enthusiast who made this possible.
A recent McKinsey report, "Gen AI’s next inflection point: From employee experimentation to organizational transformation", reveals an interesting trend. As I've posted before, when companies lack a formal adoption strategy, the benefits of workplace GenAI often go to employees rather than the business itself.
Key findings from McKinsey:
-
Employees are ahead of their companies: 91% of surveyed employees use GenAI at work, but only 13% of companies have implemented it in multiple areas.
-
Closing the gap: To turn employee enthusiasm into business value, companies should:
- Help AI-savvy employees share their knowledge with colleagues
- Start with specific business areas where GenAI can have immediate impact
- Approach GenAI adoption as a significant change management initiative, considering how it affects people's work and attitudes
-
Skills matter: Companies need to focus on training current employees rather than just hiring new talent.
-
Structure and systems: Set up clear rules for acceptable GenAI use and plan to integrate with existing processes.
While the report stresses the importance of quick action, it also makes it clear that not everyone is an early adopter. Your company still has the time to plan carefully. The goal is to move beyond individual employees trying GenAI on their own, and instead use it in ways that clearly benefit the entire business.
