ChatGPT4-Questions/User:Darwin2049/Overview

2023.11.09 - PAGE IS NOW ALL SCREWED UP....
FIRST ORDER OF BUSINESS FOR TOMORROW IS TO PUT IT BACK IN ORDER
AND FIX UP ANY OTHER DISCREPANCIES; SHOULD BE: (AND EXPLAIN WHY)
IMPRESSIONS
OPERATIONS
CAVEATS
RISKS
INTERMEDIATE OBSERVATIONS/CONCLUSIONS
PHASE SHIFT
SPECULATION
CONCLUSION

OpenAI - ChatGPT4.
In what follows we attempt to address several basic questions about the onrushing progress with the current focus of artificial intelligence. There are several competing actors in this space. These include OpenAI, DeepMind, Anthropic, and Cohere. A number of other competitors are active in the artificial intelligence market place. But for purposes of brevity and because of the overlap we will limit focus on ChatGPT4 (CG4). Further, we focus on several salient questions that that raise questions of safety, risk and prospects.
Specifically, risks that involve or are:

• Interfacing/Accessibility-Conformability - Synthesis. how will different groups interact with, respond to and be affected by it; might access modalities available to one group have positive or negative implications for other groups;
• Political/Competitive - Synthesis. how might different groups or actors gain or lose relative advantage; also, how might it be used as a tool of control;
• Evolutionary/Stratification - Synthesis. might new classifications of social categories emerge; were phenotypical bifurcations to emerge would or how would the manifest themselves;
  
• Epistemological - Synthesis how to reconcile ethical issues within a society, between societies; more specifically, might it provide solutions or results that are acceptable to the one group but unacceptable to the other group;
  

Synthesis. Four question groups (indicated above interface, political, evolution and epistemological) are the foundation upon which the analysis as presented follows. Addressing these questions meant acknowledging that they entail both explicit and implicit questions.

Game: Zero or Non-Zero Sum. Each of these question groups points to the risks associated with power shift dynamics; specifically, who stands to gain and who stands to lose as a result of this new reality;

Impressions. Addressing these question indicated that making a short survey of reporting with the intent to gauge the reported sentiment might offer some insight. The results suggested four major categories. These sentiments included (positives favor continued advances worried cautious, worried; alarmed petitioned for government intervention; battle stations! being overtaken by a rival is unacceptable);

The way that these sentiments were categorized in terms of risk involved describing the kind of risk that each seemed to be either stating explicitly or implying.

Impressions and Sentiment

Operations, i.e. might understanding CG4's internal mechanisms offer insight into how one group might gain or lose advantage? might understanding how similar Deep Learning systems work and how or if they insight into their own operational processes?
Risks. Framing the observable sentiment into categories made describing CG4 into risk categories became considerably easier. Three primary categories emerged: systemic, malicious and theoretical. The category of theoretical obliged the recognition that CG4 is inherently dual-use. The result was the category of risks that are (systemic: innate - simply because the technology exists; malicious actors: old tasks new tools; theoretical (speculative): possible now, like controlling fire);

Intermediate Synthesis Based upon what we have observed and

• Epistemological/Ethical Relativism - Synthesis how to reconcile ethical issues within a society, between societies; more specifically, might it provide solutions or results that are acceptable to the one group but unacceptable to the other group;
  

Synthesis.
These four question groups (interface, political, evolution and epistemological) form the core of what follows. Attempting to respond to these questions obliged the development of a review and response plan. What followed is a set of steps that have attempted to shed light on each of these areas; over the course of this examination the reality has been that addressing these four groups was not a straightforward one to one mapping; Instead, addressing them has meant gaining focus on informed perceptions by knowledge able observers; from there it was possible to proceed to try to answer the implicit questions relating to risk; in order to understand risk entailed getting a better picture of what underpins the CG4 technology and by corollary its peer architectures; by way of identifying and articulating the various risk categories only a short step was indicated to ask the question of "what happens now?"; this is presented in a speculative step into a near future when new tools and processes have become accepted Therefore this examination proceeds along the following lines:

• overview informed sentiment: if possible group together into camps which turned out to reduce to: (1) press on, 2) proceed with caution, 3) severe storm warnings ahead, 4) damn the torpedoes! full speed ahead!. );
• we know that they are there: knowledgeable observers have provided numerous reports of CCP hacking and related malicious efforts;
• they have signaled their intentions CCP literature and propaganda has made very clear that the West is an obstacle and that the US is the prime target;
• their approach has been adversarial: they have used clearly hostile practices to gain strategic advantage
• expert consensus confirms this reality: the PRC has clearly indicated its intent to become the global leader;
• the risk can not be ignored: for the West to do otherwise risks experiencing a potentially crippling setback; the extreme risk category asserts that maintaining the technological lead is non-negotiable;

By of proceeding, we:

• spotlight the main AI approaches: show several comparable technologies (generative, transformers, bidirectional, bidirectional encoders, generative adversarial, convolutional, recursive, large language models.);
  

• focus risk questions: the four question groups mentioned above implicitly suggest risk; the following analysis offers several examples that suggest how CG4 class systems will extend the realities of risk; systemic, malicious and theoretical); in the case of theoretical we would add that caveats apply when reviewing this group because of the inherent dual use capabilities that CG4 offers;

• exemplify risk categories: show examples of how the new technology will impact existing risks and suggest novel new risks that may arise;
  

Looking Forward. The picture that appears to be emerging suggests that looking forward a short distance in time suggests that a range of surprises await those who are using this technology. Black Swan Event. Recent developments in the area of quantum computing suggests that as this technology gains it grounding everything done with computing and especially artificial intelligence will be totally eclipsed by the capabilities that this new environment will offer.

• Impressions that have been expressed by knowledgeable observers; a short sampling of impressions and perceptions include voices that are or are saying:

• Caveats.
  After assessing the range of sentiment from very well informed observers and experts our conclusion is that a summary of caveats and qualifications might help in constraining and focusing the discussion going forward.

• An inflection point either has or momentarily will be or has been reached;
• mitigating surprise the concerned community would be well advised to regularly monitor the archive for papers so as to mitigate the gain early warning;
• Imaginative expectation is advised because novel combinations are emerging that will herald entirely new forms of CP capabilities; most of which can not be guessed at as of this writing (Fall 2023);
• quantum computing will eclipse all advances to date As of this report the major actors in this space have already reported advances. In the case of the IBM corporation their Quantum System Two may very well shatter all expectations. Current indications are that deep learning tools can be hosted on these new environments. At that point in time the ability of a deep learning CP based tool will take on qualities that can not now be estimated. Current training times have been reported to have required weeks to months to train as was the case of CG4. These times may now collapse to seconds or less.
• pseudo-AGI (PAGI) might be a way to envision what to expect. This terminology is intended to suggest that the term AGI might be too simplistic and may require further analysis and refinement.
  • Quantum Based CP (QBCP)soon-to-be formally announced quantum computing systems will become the next logical environment in which to host CPs; when this happens expect training volumes to explode and training times to collapse; incorporation of knew knowledge will become almost instant;
  • intelligent interfacing tools incorporating advanced theory of mind metrics will minimize the interval between end user motivation hypothesis and action plan to mere instants; such a capability might appear to even an expert user like interacting with an extremely well versed, empathetic and supportive colleague;
  • autonomous QBCP/PAGI linking' multiple QBCP's will form ad hoc or permanent communications pathways between themselves; these assemblages will enable the merging of multiple skills and capabilities that can not now be guessed at; these new capabilities will extend and then supersede the current trend of using communications tools such as WebEx or Zoom for conferencing; rather individuals who are in possession of their own targeted CP will enable access to peer CP's; this access might be transitory and transactional or might form the basis of new enterprises;
  • exponential growth - Cambrian explosion no capabilities such as this have yet been reported; yet there appear to be no obstacles to their development; the rise of
  • dawn of advanced composite CP's construction of modern cruise ships might serve as a basis upon which to understand the scale of capability that will shortly emerge;
  • Darwinian evolution AGI may have yet to arrive; however close approximations may soon become evident; we could see an accelerated Moore's Law of evolution where problems that have heretofore taken months or years to solve might not only be solved in a morning but that highly sophisticated access and control tools might become commonly available mid-afternoon;
• Terms, Theory of Operation

• • concepts are presented that contrast large language models with generative and generative adversarial models;
  • several other deep learning models are summarized as well for the purpose of further context;
  • LLM operation is sketched out in this section; several helpful video segments are presented that are intended to further facilitate why the LLM approach has become so successful;
  • novel and unexpected forms of emergent behavior are mentioned;
    
• Risks our approach has been to present a few commonly occurring risks these have collected into groups that are
  • organic or otherwise systemic and innate within the science or technology itself;
  • malicious actors should be expected to formulate highly elaborate netcentric attacks; we should expect that elaborate collections of actions, bait and switch, false flag and seemingly random but covertly connected acts of sabotage to become the new normal;
  • theoretical risks are speculative; but are based upon what is know known and are intended to spotlight what is possible within the purview of known capabilities;
• Intermediate Considerations.

Back to root page: ChatGPT4-Questions

Original Questions. Formulating a response to these questions took the approach of scanning whatever news reports seemed to make mention of the topic; from there look for sentiment patterns or groupings;
as this process progressed Sentiment. In order to maintain focus and forward momentum sentiment has been expressed as voices arguing in favor of further development. Other notable voices have urged caution going forward. Not least among them were such notable figures as Elon Musk. Voices expressing the same sentiment but with greater force suggested that all efforts going forward with this research should be paused for a limited amount of time. Then there were voices that expressed grave concern that this represents an existential all-of-national-resources imperative to go pedal to the metal. The risk to slow down or pause while avowed rivals are pressing ahead was just too great. In short the main camps that seem to have coalesced include: press on, proceed with caution, severe storm warnings ahead, damn the torpedoes full speed ahead.
Literature Review. Therefore the path of progress proceeded along the lines of: what are informed people saying about this new capability was foretelling. Reporting seemed to be increasing by the day. This necessitated limiting the choices of what to view and include since trying to gather it all could become a research project in and of itself. Based upon these sentiment it should come as no surprise that the risks are real and pressing. Anyone viewing a real time world map of malicious attacks shows a planet awash in twenty-four seven activity.
Model Variations. By this point in the examination it became evident that putting a spotlight the underlying technology might help. This revealed that there were several variations on the underlying approach of Deep Learning systems. Prominent ones included: generative, transformers, bidirectional, bidirectional encoders, generative adversarial, convolutional, recursive, large language models.
We briefly touch on each of these but the main focus has been on the LLM approach that uses a Pretrained approach, hence the name GPT... generative pretrained transformers.
Risks. Therefore the immediate follow on question became given that this is a major advance in a known technology how might it be incorporated, the answer to that rhetorical question fell right out into plain view as being how might this new technology pose risks, and what kinds of risks. These immediately fell into three main categories... i.e. risks that were organic (they arose innately as a result of their existence), malicious (actors with malevolent intent) and theoretical (possible new developments that are possible or newly enabled).