AGI: Beyond the Hype, Before the Storm

Visited another AI meetup on promoting Estonia as AI nation. After opening words from Vattan PS and Allan Martinson:

First talk from Linnar Viik about usage of OpenAI / Anthropic in education. I support it, AI is great at having student-tailored approach. There are however 2 dangers I see - model bias/ hallucinations in models and using closed API models as providers.

I think Estonian schools should use locally hosted models that are trained in collaboration with Estonian Ministry of Education and maybe Clarifai could help with that.

Second set of talks, was about local AI startups and their 5-min pitches from some of them: Askel.ai, Supersimple, Targum, Align (getalign.com), Forgemaster AI. I like Askel a lot, because its about agents and I'm very optimistic and bullish on that, using and building some myself.

I asked bunch of scary questions about agents and how they are going to distrupt their business, so Vattan banned me from scaring the panel speakers later on 😅

Finally, a panel discussion "AGI: Beyond the Hype, Before the Storm".

My questions for panelist would have been:

• Gert Jervan. Education institutions. Why do we need them if AI is the best teacher? Social norm? Guidance on the progression? Culture? Socialization?
• Andre Visse: Security in telecom infrastructure in AI era. If AI agent can hack a router, how is Telia going to defend itself?
• Taivo Pungas: Agents in retail on consumer side. Are we going to get them? Can they negotiate prices if culture is using fixed prices?
• Taivo Pungas: Intelligence is becoming cheap. Is everyone going to become smarter or dumber due to that?
• Vattan PS: Why should Estonia be special if AI will be accessible in all countries? Shouldn't this be the big equalizer, like open education (Courseram Harvard lectures) should have been affecting the developing world?

There were also two notable questions:

• Astra Tikas on ethics vs commerce (if I heard correctly)
• on AI heads race of domination

Afterwards, I had a debate on whether agents give you 10x productivity boost, wether we have achieved AGI, are models are better at reasoning that average human, whats going to happen next. Was surprised people don't see how agents are going to affect most of businesses.

• We are in process of making models stable enough to be more trusted using multiagentic systems
• I think its fair to ask questions like "Why do you need UI for SaaS if agents can interact with you over API"?

I think we have AGI already and models already exceed 130 IQ and we are going to get super intelligence soon, because model training will enter into self-improvement phase (that CEO of Anthropic was warning about that we need to understand internal LLM architectures at test time more).

Its a question who will make it agentic, distributed and viral and release it in.. Spain. But AI will also save life and drastically multiply human abilities.

Summary of panel talks

Here is a summary of the panel "AGI: Beyond the Hype, Before the Storm", listing the most important ideas grouped by person Disclamer: recorded with superwhisper, summarized by gemini, correctness not guaranteed

Gert Jervan, Dean, TalTech University

• AI/AGI is expected to have a major positive impact on the educational sector, enabling individualized learning and improving the process.
• AGI presents both challenges (powerful tools) and opportunities (directing beneficial processes).
• Estonia has a significant opportunity due to its high level of digitalization, data availability, small size allowing for experiments/demonstrators, and an established IT sector
• TalTech is launching an AI Systems program to train AI engineers, contributing to the necessary workforce that can use, understand, and develop AI.
• The role of teachers will evolve into mentors or reflectors, focusing on educating students to think rather than just recall facts.
• AI should be viewed as an important tool.

Andre Visse, CEO, Telia Eesti

• For infrastructure companies/vital service providers, AI/AGI brings significant risks that need to be addressed alongside opportunities.
• Infrastructure needs to support autonomous systems, which raises questions about investment, scaling, and who will pay for these capabilities.
• Automation, while beneficial, introduces complexity in troubleshooting and maintenance compared to human error (illustrated by network script incidents).
• There is a challenge in determining the safe level of autonomy for critical infrastructure and how to keep humans effectively in the loop for oversight and intervention.
• EU regulations (like the AI Act) can sometimes be slow, potentially hindering progress compared to other regions.
• AI is a tool, but users must remain in control and be aware that it can sometimes present misinformation or "hypnotize" users into believing its output. It does not "think" or have personality.

Taivo Pungas, CTO, Pactum

• Based on practical experience, it feels like "jagged AGI" already exists – superhuman in specific tasks but useless for basic physical tasks.
• Provides personal examples of using AI for complex research, meeting preparation, scripting, and evaluating purchasing decisions, highlighting its current usefulness.
• As intelligence becomes cheap and abundant, the compliments to AI become more valuable (basic economic principle).
• Key complements to AI include soft skills, agency (AI won't initiate projects), selecting the right problems/tasks, discerning usefulness, general management, and providing clear instructions/goals.
• Robot dexterity is currently limited.

Vattan

• AI "agents" follow protocols, use loops, and leverage external APIs (or "MCP") to access information beyond their training data.
• AGI can be viewed as a potential business model.
• Estonia needs an "AI Mafia" (influential network) willing to support young talent and entrepreneurs to foster breakthrough innovation.
• Direct funding for AI research and development is needed, not just bundled under other sectors.
• Significant opportunities for Estonia lie in AI security (needed as AI grows) and E-governance (AI requires governance, and Estonia has an advantage here).
• Training skilled AI/ML engineers is a difficult and time-consuming process.
• Suggests building a federated/predicted learning layer on top of Estonia's X-Road system (e.g., for health data) as a concrete innovation opportunity.

Lembit Loo

• The exact definition of AGI is less important than the fact that we are approaching significant capabilities ("at the finish line").
• Estonia needs to figure out how to effectively utilize the AGI wave through cooperation and partnerships.
• Estonia has a unique position: capable of understanding/using technology (like the US and China) while also being part of the EU regulatory framework (like the AI Act).