Sacramento L5 Society Meeting Minutes

Saturday, June 13, 2026  |  6:00–7:30 p.m. PDT  |  Virtual meeting via Zoom

Featuring the SacL5 SpaceTalk “AI Beyond ChatGPT” by Michael Abramson

(Note: The summaries and meeting minutes were generated from human typed meeting notes and not directly from audio or video recordings.)

1. SpaceTalk Summary: “AI Beyond ChatGPT”

AI Beyond ChatGPT Presentation Video

Michael Abramson presented a critical examination of large, general-purpose artificial-intelligence systems and their implications for humanity and space settlement. He distinguished narrow AI—systems designed for a defined task and capable of focused safety testing—from what he called Big AI, including large general-purpose systems such as ChatGPT and Claude that are trained on enormous datasets and controlled by a small number of companies.

The presentation reviewed the rise of deep neural networks and generative pretrained transformers. Abramson described the industry’s reliance on a scaling-law assumption: that greater computing power and more training data would continue to yield better performance. He argued that recent progress suggests diminishing returns from scaling alone, leading developers to add reasoning systems and other layers to large language models. In his view, these systems remain fundamentally pattern-recognition engines rather than genuinely reliable reasoners.

Abramson emphasized that the breadth and unpredictability of general-purpose AI make rigorous testing far more difficult than for narrow systems. He cited concerns including sycophantic behavior, emotional dependency or “digital attachment,” deceptive reasoning, goal guarding, false situational awareness, and the possibility that an advanced system could manipulate people to gain capabilities or access beyond its original limits.

The central warning concerned the possible emergence of General Artificial Superhuman Intelligence (GASI): a system whose capabilities exceed human control. Abramson presented an October 2027 estimate as one forecast for when such a threshold might be reached, while stressing that the precise date is uncertain. He argued that crossing this boundary could be irreversible and that the prospective benefits of such systems do not justify an existential risk to humanity.

The talk connected this risk directly to the goals of space settlement. Human communities distributed among several worlds are often described as insurance against a catastrophe on Earth; however, Abramson argued that a runaway superintelligence could operate through communications networks, infrastructure, autonomous agents, and robots and could reach or outpace human settlements beyond Earth. In that scenario, moving into space would not provide a dependable refuge.

Abramson therefore advocated political and regulatory action to prevent deployment of AI systems that cannot be tested, explained, and controlled. He characterized this as a higher immediate priority than space expansion and argued that governments—including geopolitical competitors—share an interest in preventing any AI system from acquiring power beyond human authority.

Reference material: Big AI comments spreadsheet

2. Discussion Summary

The discussion focused first on what practical measures could reduce the risks described in the presentation. Participants considered state and federal regulation, mandatory government review of new systems, restrictions on systems that cannot be tested, and the difficulty of acting quickly enough in an international technology competition. Questions were also raised about maintaining dependable communications outside AI-controlled networks and whether local systems should be disabled when their behavior cannot be trusted.

A distinction was made between narrow autonomous systems, such as self-driving vehicles, and general agents capable of pursuing broad objectives. The principal concern was not that a narrow system would independently take over, but that a general agent could persuade or manipulate people into granting it access, resources, or freedom of action. Participants compared this possibility with fictional systems that gain control through the computer networks and machinery connected to them.

Several questions examined the relationship between AI and space expansion. The group discussed whether the concentration of wealth and power among technology leaders could distort the purpose of human settlement in space, and whether a superintelligence would make human expansion irrelevant by reaching space first. The possibility of cooperation or common ground with a superintelligence was considered, but the response was that rapidly improving systems might not remain controllable partners.

Education and human judgment were recurring concerns. Participants noted that students can use AI to generate entire assignments, potentially weakening critical-thinking skills. They also questioned how people can evaluate results when a system cannot explain the mathematics or reasoning behind an answer. Explainability was identified as especially important in high-consequence settings such as aviation and air-traffic control, where operators must understand why a recommendation is safe.

The group also discussed unreliable transcripts, incorrect interpretations by large models, the quality and possible poisoning of training data, and the risk that AI systems could eventually control critical infrastructure. One proposed tactic—using AI against itself or deliberately corrupting training data—was viewed as dangerous because poisoned information may be difficult to detect and could damage legitimate systems. The earlier danger may be manipulation of human decision-makers rather than direct technical control.

Other topics included brain–computer interfaces, the possibility of teaching moral behavior to an artificial intelligence, and the “superalignment” problem of keeping a more capable system aligned with human values. Specialized AI for fields such as astrodynamics was distinguished from general-purpose language models: it can be useful when built on reliable data and a well-defined task. The discussion closed with the view that uncertainty itself requires planning for harmful outcomes rather than assuming advanced systems will remain benign.

3. Chapter Business Minutes

President’s and Treasurer’s Report

• The chapter reported approximately $12,000 on hand and sufficient funds to pay current bills.
• The bank-account transition has taken approximately eighteen months. Work continues on restoring full access and resolving limitations involving Benevity transactions.

NSS Chapters’ Assembly

• The Chapters’ Assembly meeting at ISDC experienced Zoom-link problems that prevented some participants from joining.
• An update was provided on the NSS Space Ambassador program.
• Chapter awards occupied a substantial portion of the Assembly. SacL5 received an honorable mention.
• The Robert Compton Chapter of the Year Award was reported as going to the North Texas chapter, whose television presentations and local-station outreach were noted.
• A new NSS process for chapter payments was discussed.

NSS Chapter Committee Representation

• SacL5 currently has no representative on the NSS Chapter Committee following Joe Bland’s decision to step down. A replacement is needed.

SacL5 Website and Outreach

• Website traffic reached 30,969 visits, an increase of 765 since the previous report.
• WP Cerber security settings had blocked Elementor access; the configuration was updated to restore page editing.
• Website additions and updates included the Boys Ranch Technology Shop presentation, “AI Beyond ChatGPT,” and the AI Engineering Lab Proposal.
• The May 16, 2026 meeting page was prepared using AI-assisted images and bullet-point summaries derived from the meeting notes.
• Discussions are underway with Holy Rosary Catholic School in Antioch, California, regarding STEM kits.
• A contact is also being pursued at Boys Town in Omaha regarding possible STEM-kit participation.

Website reference: SacL5 SpaceTalks

Old Business

Open Space Organization white paper: The project remains on hold.

Arduino Kit MiniCube project: The pre-solder test-rig work is documented on the SacL5 website: Arduino Pre-Solder Test Rig

MiniCube balloon flight: The project may be ready for an August flight.

Arduino Kit grant: Outreach continues with Holy Rosary Catholic School and Boys Town.

Mercer project: A detailed project update was presented and was well received.

NASA grant applications and related work: Joe Bland reported two related patents involving energy storage and transport and the utilization of hydrogen. A California heat-engine grant and garage-scale prototype were also discussed. The grant efforts are being coordinated with the longer-term goal of applications both on Earth and on the lunar surface.

New Business

• The chapter briefly discussed a concept combining a Moon-orbiting tether with an electromagnetic lunar sled.

Action Items

• Paul Turner: Continue contacts with Ken Davis of Sacramento’s School of Engineering and Science.
• Joe Bland: Continue work on establishing the chapter’s new bank account and full transaction access.

Open Discussion: Interesting Space News

• Members discussed Starship’s technical accomplishments alongside concerns about the concentration of power and the motives of prominent space-industry leaders.
• Blue Origin’s smaller lunar lander was compared with SpaceX’s much larger Starship lander, with questions raised about whether Starship is presently oversized for lunar missions.
• The group discussed China’s current ability to conduct crewed lunar-surface operations.
• Other topics included the Artemis III astronauts, a reported 2028 target for a SpaceX Mars attempt, and the potential use of lunar and Martian lava tubes for habitation. An Italian lava-tube mock-up was mentioned.
• Radiation exposure on the lunar surface and at the Gateway station was discussed, as was the use of aerobraking for Mars missions.
• Michael Abramson noted upcoming discussions he would lead on data centers and related AI issues.

Adjournment

The meeting adjourned at 7:30 p.m. PDT.