Victor Senkevich (@victorsenkevich) 's Twitter Profile
Victor Senkevich

@victorsenkevich

🔸AGI Guru. CTO.
🔸AGI is closer than it appears in the mirror of our perception.
🔸The problem is solvable if it is codable.

ID: 252269567

calendar_today14-02-2011 21:13:47

1,1K Tweet

79 Takipçi

33 Takip Edilen

Victor Senkevich (@victorsenkevich) 's Twitter Profile Photo

• GenAI / LLM is a "predictive" AI • AGI is a "predicative" AI, capable of logical reasoning by definition, i.e. processing predicate logic  • Feel the difference • "Predicting" will never replace understanding and reasoning

• GenAI / LLM is a "predictive" AI 
• AGI is a "predicative" AI, capable of logical reasoning by definition, i.e. processing predicate logic  
• Feel the difference
• "Predicting" will never replace understanding and reasoning
Victor Senkevich (@victorsenkevich) 's Twitter Profile Photo

• A short formula for LLMs: Useful But Unreliable • Learning is important and "useful", but it cannot compensate the lack of understanding • This is the whole essence of LLM • Transformation from LLM to AGI is transformation from quantitative learning to qualitative reasoning

• A short formula for LLMs: Useful But Unreliable
• Learning is important and "useful", but it cannot compensate the lack of understanding
• This is the whole essence of LLM
• Transformation from LLM to AGI is transformation from quantitative learning to qualitative reasoning
Victor Senkevich (@victorsenkevich) 's Twitter Profile Photo

• Real AI forming its goals independently cannot be effectively "aligned" • But we can and we should "align" the AI motivation, partially limiting it by the rules • Agency is determined by motivation • "Motivation" is the keyword for AI alignment linkedin.com/posts/victorse…

• Real AI forming its goals independently cannot be effectively "aligned"
 • But we can and we should "align" the AI motivation, partially limiting it by the rules
• Agency is determined by motivation
• "Motivation" is the keyword for AI alignment
linkedin.com/posts/victorse…
Victor Senkevich (@victorsenkevich) 's Twitter Profile Photo

• "Embedded GenAI" is an oxymoron. However, "embedded AGI" is inevitable. Because: ▲ Thinking isn't computing ©VS ...although computing is used in thinking. linkedin.com/posts/victorse…

• "Embedded GenAI" is an oxymoron. However, "embedded AGI" is inevitable. Because:
 ▲ Thinking isn't computing ©VS
...although computing is used in thinking.
linkedin.com/posts/victorse…
Victor Senkevich (@victorsenkevich) 's Twitter Profile Photo

• LLMs are unable to perceive the structure of the text/document due to their probabilistic nature • Actually, LLMs need clean, well-structured, semantically enriched data to provide inaccurate and unreliable results 😎 linkedin.com/posts/victorse…

• LLMs are unable to perceive the structure of the text/document due to their probabilistic nature
• Actually, LLMs need clean, well-structured, semantically enriched data to provide inaccurate and unreliable results 😎
linkedin.com/posts/victorse…
Victor Senkevich (@victorsenkevich) 's Twitter Profile Photo

Here is my simplest AGI formula: ▲ AGI = R + R (Recognition + Reasoning) ▲ AGI Agent = R + R + M (Recognition + Reasoning + Motivation) Definitions: ▲ Agency is a predetermined subjectness ©VS ▲ A subject is an object with motivation ©VS lnkd.in/dtt3w74b

Here is my simplest AGI formula: 
▲ AGI = R + R (Recognition + Reasoning) 
▲ AGI Agent = R + R + M (Recognition + Reasoning + Motivation)  

Definitions: 
▲ Agency is a predetermined subjectness ©VS 
▲ A subject is an object with motivation ©VS lnkd.in/dtt3w74b
Victor Senkevich (@victorsenkevich) 's Twitter Profile Photo

📖"Why Language Models Hallucinate" / OpenAI ❌Unfortunately, the article doesn't specify the main cause of "hallucinations" and the proposed method to fix the problem is ineffective 👉The reason for LLM's "hallucinations" is simple: semantic domain mixing lnkd.in/gBmytpPr

Victor Senkevich (@victorsenkevich) 's Twitter Profile Photo

• Any high-quality data are useless if the data-to-knowledge algorithm is ineffective • Why is data quality the AI’s biggest challenge? I'm afraid this question needs to be reformulated. Why are even high-quality data often become an AI's challenge? lnkd.in/gSyHuXdA

• Any high-quality data are useless if the data-to-knowledge algorithm is ineffective
• Why is data quality the AI’s biggest challenge? I'm afraid this question needs to be reformulated. Why are even high-quality data often become an AI's challenge?
lnkd.in/gSyHuXdA
Victor Senkevich (@victorsenkevich) 's Twitter Profile Photo

• The proper domain configuration initially distinguishes potentially successful GenAI applications from obviously unsuccessful ones. lnkd.in/gw829Djx

• The proper domain configuration initially distinguishes potentially successful GenAI applications from obviously unsuccessful ones.
lnkd.in/gw829Djx
Victor Senkevich (@victorsenkevich) 's Twitter Profile Photo

👉 The difference between LLM and AGI is quite simple, essentially being the difference between the concepts of "learning" and "understanding," but the correct definitions of these concepts require rigorous formalization. lnkd.in/g-jYp-gp

👉 The difference between LLM and AGI is quite simple, essentially being the difference between the concepts of "learning" and "understanding," but the correct definitions of these concepts require rigorous formalization.
lnkd.in/g-jYp-gp
Victor Senkevich (@victorsenkevich) 's Twitter Profile Photo

❓ What does LLM determinism mean? 👉 Previously non-deterministic "hallucinations" will now be deterministic, meaning they will repeat with high probability 😎 ✋ The "determinism" of LLM answers does not negate its probabilistic nature lnkd.in/gipCeUET

❓ What does LLM determinism mean?
👉 Previously non-deterministic "hallucinations" will now be deterministic, meaning they will repeat with high probability 😎
✋ The "determinism" of LLM answers does not negate its probabilistic nature
lnkd.in/gipCeUET
Victor Senkevich (@victorsenkevich) 's Twitter Profile Photo

• Consciousness is, basically, a simple thing, if we don't multiply unnecessary entities (remembering Occam 🙏). • The shortest definition: ▲Consciousness is perception with understanding ⚬ I.e. consciousness is a meaningful perception. lnkd.in/gcZCdwDA

• Consciousness is, basically, a simple thing, if we don't multiply unnecessary entities (remembering Occam 🙏). 
• The shortest definition: 
▲Consciousness is perception with understanding
⚬ I.e. consciousness is a meaningful perception.
lnkd.in/gcZCdwDA
Victor Senkevich (@victorsenkevich) 's Twitter Profile Photo

• Now is the best time to shift the focus from Data to Knowledge. • Enterprise AI isn't Data-driven, it's Knowledge-driven. • Feel the difference between atomised Data and aggregated Knowledge. lnkd.in/gqik7K2U

• Now is the best time to shift the focus from Data to Knowledge. 
• Enterprise AI isn't Data-driven, it's Knowledge-driven.
• Feel the difference between atomised Data and aggregated Knowledge.
lnkd.in/gqik7K2U
Victor Senkevich (@victorsenkevich) 's Twitter Profile Photo

• Unclear / poorly defined AI Mental Domain (AIMeD) inevitably leads to erroneous AI inference, REGARDLESS of the data quality • Technologies either work or they are discarded • Enterprise AI Should Be Properly AIMeD ©VS • Data Alone Is Not Enough lnkd.in/guwY6vMU

• Unclear / poorly defined AI Mental Domain (AIMeD) inevitably leads to erroneous AI inference, REGARDLESS of the data quality
• Technologies either work or they are discarded
• Enterprise AI Should Be Properly AIMeD ©VS
• Data Alone Is Not Enough
lnkd.in/guwY6vMU
Victor Senkevich (@victorsenkevich) 's Twitter Profile Photo

• Instead of limiting the freedom of each cow in the herd / restrict inference for each AI instance, we should fence off the entire pasture / rigorously predetermine the entire AI Mental Domain (AIMeD) / Knowledge Scope. • AI Safety: AIMeD Approach: lnkd.in/gNRi5Hiu

• Instead of limiting the freedom of each cow in the herd / restrict inference for each AI instance, we should fence off the entire pasture / rigorously predetermine the entire AI Mental Domain (AIMeD) / Knowledge Scope.
• AI Safety: AIMeD Approach:
lnkd.in/gNRi5Hiu
Victor Senkevich (@victorsenkevich) 's Twitter Profile Photo

Adaptive Enterprise Knowledge Graph / AEKG is a hard-coded set of nodes representing actions (procedure calls, database searches, user interactions etc.) and a set of edges representing action selection algorithms that can be dynamically generated... lnkd.in/gVPv7xjh

Adaptive Enterprise Knowledge Graph / AEKG is a hard-coded set of nodes representing actions (procedure calls, database searches, user interactions etc.) and a set of edges representing action selection algorithms that can be dynamically generated...
lnkd.in/gVPv7xjh
Victor Senkevich (@victorsenkevich) 's Twitter Profile Photo

• #EnterpriseAI successful project keywords: "reliability" (1) and "integration" (2). • #OpenAI has failed to implement a #ChatGPT at #Walmart for a simple reason: They don't have the first one by definition, and they haven't provided the second one lnkd.in/ejVjFcXT

• #EnterpriseAI successful project keywords: "reliability" (1) and "integration" (2).
• #OpenAI has failed to implement a #ChatGPT at #Walmart for a simple reason:
They don't have the first one by definition, and they haven't provided the second one
lnkd.in/ejVjFcXT