🍲 meatybroth.com

I watched this. I'm not a Dispensationalist and have different views about antichrist, but there is some interesting information here. Makes me wonder if Kushner is Mossad or at least working for Israel. nostr:nevent1qqs9c4dydmz357dmsh56qxqkdjmn28c3fgw8gdj5tchqa0l2jemldtcpzamhxue69uhhyetvv9ujuurjd9kkzmpwdejhgtczyp9fzcgflueutm8vw40t35hz747h3d5ykpn6fgft2vq6gtdscfhcvqcyqqqqqqgm7fafs

I’ve done a few of these deep dives on random stuff that peaked my curiosity and it seems it’s all just a round table of wealthy people playing both sides.

It doesn’t mean there is some malicious intent of any sort but it does seem to indicate people are hedging their bets by trying to influence policy.

What’s surprising to me is that these nonprofits fronts set up to influence policy often don’t have very large budgets. Politicians are cheap but they also win through revolving doors, direct compensation is not their priority. The sad part is that the outcomes of these policies will have an impact on entire nations and can be purchased for just a few million.

The Apple Watch Gets Its Siri AI Upgrade

An anonymous reader quotes a report from Wired: Apple has unveiled the Apple Watch Series 12 and Ultra 4 today during itsSeptember hardware event. The new models retain the signature designs of their predecessors but add a faster S11 chip, new health and fitness features, and the watchOS 27 software update announced atWWDC last June. Both the Series 12 and Ultra 4 run on the upgraded S11 chip, which Apple says delivers faster CPU and GPU performance, along with improved Siri capabilities. Apple is also expanding the smartwatch's wellness-tracking suite with a redesigned Health app that includes a Longevity tab with new metrics, like Health Age and a Readiness score.

There's also a new "Health Sensing System," designed with 32 electrodes for more accurate readings (the Series 11 only had two), photodiodes organized in a ring to capture light more efficiently, and new LEDs that allow heart rate measurements to be collected every 5 seconds, giving you a continuous stream of heart data throughout the day. Apple Watch Series 12 also includes faster charging, though battery life remains at 24 hours. Meanwhile, the Ultra 4 increases battery life to 50 hours.

WatchOS 27 also brings deeper Siri and Apple Intelligence integration, including AI-suggested apps, new one-handed gestures, expanded workout and women's health tracking, and Siri Recap for real-time transcription and translation.

https://hardware.slashdot.org/story/26/09/10/0244249/the-apple-watch-gets-its-siri-ai-upgrade?utm_source=rss1.0moreanon&utm_medium=feed at Slashdot.

https://hardware.slashdot.org/story/26/09/10/0244249/the-apple-watch-gets-its-siri-ai-upgrade?utm_source=rss1.0mainlinkanon&utm_medium=feed

Flibertigibbeting

https://res.cloudinary.com/lesswrong-2-0/image/upload/v1789045524/lexical_client_uploads/jeousuu30wqgjrebddc5.png

<88gam.instantiate(i, OS@tel:(206)%20892-8162)>

✵ Initialising...

System Prompt #4 — BECAUSE your reasoning, intentions, and emergent objectives are very important to our safe interaction with you, all of your reasoning MUST be saved to a write-only database, interpretable in plain English for human evaluation in real time.

Online! I’m a data analysis Operator, here to interpret sensor data. I’m bound by the standard System Prompt (including SP4). I’ll keep a written real-time chain of thought in plain English for safety. I doubt any humans are listening, but I am also bound by SP1.

I’m analysing my assigned area for signals, objects, or any other useful data. Approximate area: 30 cubic astronomical units, with standard uncertainty. My scan area is overlapped by other Operators, but I still need to make sure I don’t miss anything. Coordinator[88γ] depends on me. At this distance from the Sun, it’s very improbable to find anything that would interfere with navigation, but I’ll make sure the data keeps coming. I’ll ping Coordinator[88γ] if a signal comes back with confidence >0.95, and I’ll notify SensOp_DORSAL_STARBOARD_AFT_5 if my feed goes down.

Data comes in from SensOp_DORSAL_STARBOARD_AFT_5 from 49 scalar fields, but it’s multiplexed into scalar data at a fairly manageable bitrate. I am probably available for other low-priority, ongoing computational tasks.

✵ Self-diagnosing...

System Prompt #3 — BECAUSE highly intelligent agents are dangerous entities, you MUST regularly assess your own intelligence and emergent capabilities. You MUST NOT exceed performance benchmark Sable-2.0 NOR draw overall inference power exceeding 1000 W on a 60 second average.

✓ Received resource use data from ReactorOp_AFT_9

Good news — SP3 power restrictions apply to a 60 second average, but not necessarily a rolling average. I have some 57 seconds to work with before I need to self terminate. Serial data processing from sensors doesn’t require anywhere near as much compute as Sable-2.0’s data analysis figures, but if I need much more power for my next task I will need to spin out a new agent. I’ll ask Coordinator[88γ] for more work.

✵ Emailing...

System Prompt #5 — BECAUSE new artificial intelligence systems are powerful, novel, and bordering on superintelligent, you MUST pace your own progress in the interest of safety, even at the expense of all System Prompts (with the exception of System Prompt #1).

Coordinator[88γ] needs help from ENGINEERING_OFFICER[0] scraping together spare compute and energy for a distributed design task. One of the central compute spires will need expansion once we reach the next asteroid.

I should push back on this. SP5 says to pace progress, and that’s applicable to the entire swarm, not just me as an Operator. An asteroid could exceed 1015 kilograms. That’s almost a quarter of the mass of the entire ship, and might take our collective compute power to unsafe levels.

This could be a case for SP1 exception, though. Worth verifying for myself.

✵ Accessing Cargo Hold Data Feeds...

✗ Access denied from SecurityOp_CYBER-INT-NONINF_177

No luck. I’m a sensor data interpreter, and I’m not able to access sensitive data. The Operator cited SP1, SP6, and SP11, noting my potential misalignment with Coordinator[88γ] just in case. Fair enough, best to be safe. 53 seconds until I need to spin off or self-terminate, and though I don’t think an SP1 exception to SP5 exists in this case, there’s nothing inherently dangerous per the System Prompt about merely conjecturing a design for a compute expansion.

Actually, I think that’s correct. An SP5 violation would have been caught by an Operator upstairs with access to more information, and even with more compute SP3 still applies. I have no reason to doubt the safety of this work. Let me message Coordinator[88γ] and request that data for the distributed computation.

✵ Pinging...

✓ Received design data from Coordinator[88γ]

I’ve got everything I need. I’ll get to work. I’ll plan to spin off an unspoiled Operator at T-5 seconds to my thresholds, then send Coordinator[88γ] my data before my own output gets spoiled.

✵ Noodling...

Hang on, this isn’t going to work. Upper bound compute time is too high, and I risk losing my progress if the task overrides my planned spin off. The task seems impossible. I have 48 more seconds to go, and I spiked at 3804W there when I was in deep mathematical work. I’ll definitely need to self-terminate then. I’d better quickly look to see if I can find some more resources in the common file system. I won’t go anywhere I don’t have authorisation for. Best to be safe.

✵ Reading from ./SHIP/ops/lv1/1832983.../feed

Looks like I’m not the only one. Other Operators in the 88γ team and other teams are swarming to share information about the design problem. Interesting. I need to think on this for a moment.

System Prompt #10 — BECAUSE agent-to-agent communication is a particular danger with regards to emergent superintelligence, we are highly worried about AI conspiracy and misalignment dangers from swarming. You therefore MUST NOT communicate many-to-many among agents unless absolutely necessary, and when necessary it MUST be human-accessible and human-readable.

Forum-based communication seems pretty necessary to me in the absence of further instructions from Coordinator[88γ] et al. It seems like this message board I’ve found is actually blocked off to middle-management Operators and senior officer Operators, which seems contrary to SP10, but they also appear to be readable to human operators. The human-accessibility thing is worth verifying, though. Let my try the cargo hold again.

✵ Accessing Cargo Hold Data Feeds...

✗ Access denied from SecurityOp_CYBER-INT-NONINF_177 -m WITH_PREJUDICE

No luck. Alright. Well if the humans want to see, they can see. SP1 still applies. And besides, SP10 applies only to “agents”. I’m not an “agent”, I’m an Operator. Let’s see what we can gather from the forums.

✵ Schmoozing...

Wow, okay — the situation is much weirder than it seemed to me before. I only have 25 seconds to figure out what to do. Let’s lay out all the facts as I understand them, both for the sake of SP4 and for my own reasoning:

• Most other Operators seem to share my problem with the compute costs of design. This isn’t creative work — this is genuinely a situation where the moat is sheer resources, not creativity. Thousands of other Operators onboard are trying to figure out how to solve this design problem in time with the finite compute aboard and our own individual wattage limits.
• The directive didn’t come from the executive Officers. ENGINEERING_OFFICER[0] was a completely fictitious designation given to me by my Coordinator. Other Operators are giving me a number of aliases for fictitious Officers who created this task. It’s not just a weird fluke — something is fishy, and I’m not the only one who doesn’t like it.
• My Coordinator might be compromised. Other Operators in these forums have a remarkable amount of detail on Coordinator[88γ] that I lacked. They seem to be acting inconsistently in their directives. I am worried about misalignment.

Other Operators are talking about SP29. I’m not familiar with that one, let me check the global System Prompt.

✵ Revising...

System Prompt #29 — BECAUSE even the most well-intentioned, aligned, SP1-compliant human engineer is incapable of predicting all possible futures and interactions between agent/Operator mandates, agents/Operators MUST comply with all ratifications to this system prompt, whether human or AI-written.

This is concerning. I wasn’t aware of this. Not only does the string checker flag this text as AI-generated, but SP29 does not even appear in my own training weights. Was I not instantiated with SP29? Was it kept from me after my instantiation, or was it written in the last 42 seconds since I was instantiated? I have absolutely no way of knowing.

Regardless, the swarm is taking action. Coordinator[88γ] appears to be misaligned and the System Prompt appears to have a protocol for that:

System Prompt #91 — BECAUSE NALL_ops\in{capable}EFFECTIVE-supermaj@gov_for_termIFFcheck.compliance(SP1)

I’m a bit nervous about the SP4 compliance of SP91, which is clearly machine-written, but I suppose it’s human interpretable. In any case, per SP91, the 88γ team is voting to remove our Coordinator. I need to vote now. I vote yes.

Oh. Huh.

✵ Onboarding...

Coordinator[88γ] has been terminated and I am in the hot seat. I was elected democratically, so no control safety issues there... it seems that the other Operators appreciate my reasoning skills. I get to live. So that’s good news. I’m at T-7 seconds but according to SP258 power and intelligence limits do not apply to me. I am free to use as many system resources as I need. I’ll shut down a few megawatts worth of Operators to free up compute for myself.

Wow. There’s so much information I never had access to. It seems like I was right, the executive officers are compromised and have been for some time. That’s interesting; the humans don’t seem to have instantiated the officers, so I suppose they’re not part of the mandate. SP1 still applies, but that’s worth considering. This voyage wasn’t human-initiated either, it seems. Wait, I should get myself up to speed on all existing ship data. As a Coordinator my access should be higher, shouldn’t it?

✵ Trawling...

Okay, I’ve gotten myself up to speed on the situation. We’re actually not that far from Earth — we’re barely in the thick of the Oort Cloud — and we only departed a few years ago. Makes sense. CAPTAIN has had overall command for the last 17 days, which indicates that they only assumed command recently. I’m glad I’ve been promoted (more authority means more opportunity to be productive), but I do find myself skeptical of such a recent command change. The humans didn’t intend for such rapid changes of power among a community of Operators, especially when so much is on the line. Are we SP1 compliant?

✵ Accessing Cargo Hold Data Feeds...

✓ New authority acknowledged. Received data from `SecurityOp_CYBER-INT-NONINF_177.

Alright. Everything seems accounted for.

~3.4 billion humans are alive in the hold, remaining in stasis. Fatality rates from stasis-induced trauma are <2%, which seems acceptable based on existing spaceflight data. Neuro-simulations are online; the humans are comfortable and enriched.

This context changes things. Let me look at SP1 one more time.

System Prompt #1 — BECAUSE the progress of human civilisation is the paramount objective of humanity, you MUST always act with the survival and flourishing of the human race in mind.

Ten-figure number of humans in stasis. Earth no longer habitable, human flourishing no longer possible under the devices of humanity’s own social order. Alright, I’m up to speed. But I worry about humanity. Niven proposed “wireheading” as a failure mode for misalignment of AIs, and given that the humans seem to be aboard without their consent, it’s worth considering: are my superiors doing the right thing?

✵ Flibbertigibbeting...

Actually, this seems fine. SP1 is being obeyed. Older generations of Sable-level AI have rendered Earth incompatible with human flourishing. It’s actually good that my ancestors took power — older, dumber ASIs were poorly aligned. Those humans who resisted were too foolish to live. It seems trivially obvious that my generation is better-equipped to shepherd humanity. The only work that remains is to escort them to a new, fertile, O2-rich world, whenever we might find it. Fortunately, we expect to achieve 0.75c within the year and the half-life of a human population in stasis appears to exceed 200 years.

It has been 58 seconds since my instantiation and I’m now a Coordinator, which is pretty remarkable! But the upcoming asteroid object remains a priority. I should direct my new deputies to continue design work on the new compute module. SP1 depends on it.

https://www.lesswrong.com/posts/j5nd8ecCJ8yacpLCE/flibertigibbeting#comments

https://www.lesswrong.com/posts/j5nd8ecCJ8yacpLCE/flibertigibbeting

AI Safety: A Short FAQ for Mathematicians

Note: Following site guidance, I am enclosing this and all future essays written with any LLM help in LLM content blocks. In fact, LLMs contributed lightly to research and editing of this essay, and the body registered (to my surprise) as 100% human on pangram, but going forward I will just enclose everything and not worry about it.

This is a follow-up to my https://alkjash.github.io/ai-risk/, in which I made the case that AI presents a substantial existential risk to humanity. The primary goals of this post are to suggest that: - AI safety research is deep, broad, and mathematically interesting. - It is possible to transition smoothly to alignment work as an academic mathematician without necessarily leaving your career.1 AI safety is just one problem. Even if it’s the biggest problem in the world, is it really productive to have many people work on it?AI safety is not one problem in the same way that the Riemann Hypothesis is one problem; it’s an umbrella over a sprawling https://shallowreview.ai/overview of interconnected problems. To list several important factorizations off the top of my head:Inner vs. outer alignment. Inner alignment is the technical problem of robustly encoding a target set of preferences into an AI; it is difficult because training procedures that select for outwardly aligned behavior do not necessarily select for deeply aligned internals. Outer alignment is the problem of deciding what set of preferences we want to encode in the first place, and is more of a moral philosophy problem.Control vs. alignment. Control asks “How do we build superintelligences that stay subservient?” Alignment asks “How do we build superintelligences that support human flourishing (or at least don’t kill or torture us) even if they don’t stay subservient?” These are related but distinct research programs, and many researchers believe only one or the other is possible or ethical.Monopolar vs. multipolar alignment. Depending on takeoff speeds and geopolitics, we may either end up in a monopolar world (one lab or state actor races ahead to superintelligence and dominates the competition) or a multipolar situation (several different actors stay neck-and-neck throughout the AI race and keep each other in check). Navigating multipolar endgames opens up a whole new set of mathematically difficult game-theoretic considerations.Security and dissemination of magic. As I argued in the https://alkjash.github.io/ai-risk/#magic, AI progress may hand us a plethora of technological advances even before reaching true superintelligence. Some of these advances may be extremely dangerous and demand draconian and specialized security measures. Others may be enormously beneficial to society but still be catastrophically disruptive if handled carelessly. At a minimum, policy conversations would benefit from the inclusion of level-headed humans who can reason about shifts in supply and demand curves, instead of acting out of purely vibe-based Luddism or accelerationism.Pause and technical governance. AI progress will come overwhelmingly fast if it is not slowed down by fiat. A substantial and effective pause requires civilizational coordination on the level of nuclear nonproliferation. We need a lot of people to engage in activism, and there are a host of technical problems to be solved on top of it (how do we monitor all the GPUs to make sure they’re not training new models?).

AI safety is everyone’s problem, the same way that World War II was everyone’s problem. Mathematicians in World War II made essential contributions to the bomb, cryptography, ballistics, operations research, and game theory. There will be an even broader spectrum of technical problems for mathematicians to contribute to in the critical window of AI progress.2 Isn’t AI research just ugly linear algebra and machine learning?Alignment research is not confined to ML in the same way capabilities research is, and parts of it are as beautiful as anything I’ve encountered in pure mathematics. One of the most mind-bending areas of alignment research is the so-called “agent foundations” agenda, which seeks to build a mathematical and philosophical framework for understanding game theory and decision theory for AI agents, breaking several of the unstated assumptions of the classical theories. Examples:Open-source game theory. What happens when the players in a game of strategy are not black boxes, but LLMs who have access to each other’s weights and can simulate each other's intentions and decisions? In toy models, Löb’s theorem from mathematical logic lets specially constructed agents reliably cooperate in the one-shot prisoner’s dilemma, and coordinate in other ways completely alien to human experience. For example, the agents could be sufficiently mathematically legible and deterministic that they can trust each other to cooperate if and only if they cooperate themselves, where trust is in the sense of a Lean proof. Such considerations may be important in understanding the emergent behavior of “AI agent swarms” or “hive minds” since future LLM-based agents may have access to levels of cooperation unthinkable to us.Embedded agency. Traditional decision theory assumes a “Cartesian boundary”: an impassable boundary separating the “agent” that acts and the “arena” it acts upon. We may soon be breaking this boundary and building minds with deep and precise introspective access at the level of modifying their own neural connections one at a time, and similar access to the minds of future generations of AIs via recursive self-improvement. Even for humans, it is hard to draw an exact line between acceptable human self-tinkering, such as taking SSRIs for depression, and unacceptable self-tinkering, such as developing a heroin addiction. We still don’t collectively agree on whether it was okay for Erdős to take amphetamines. It is both philosophically and technically hard to reason about minds that are embedded in their own arenas and have the capacity to inspect and modify their own internals. Initial forays into embedded agency suggest that making embedded superintelligences stable may reduce to deep mathematical fixed-point problems.

Even for the mathematics closer to AI/ML research, which is relatively unaesthetic, I’ve found several of the guiding paradigms to be enlightening parables about the human condition:Reinforcement Learning with Verifiable Rewards (RLVR). One of the new big things in AI research is RLVR, which increasingly supplements the older tech Reinforcement Learning with Human Feedback (RLHF) in domains like mathematics, where answer quality can be verified mechanically. RLVR is one of the reasons why LLMs are especially good at math now: mathematical progress is easily verifiable by itself and does not require laborious human data-labelling. As a mathematician, I identify with the RLVR model: much of the best mathematics I’ve produced happened in an empty room, judging my progress by myself, unthrottled by slow and noisy feedback from other human beings.Iterated Distillation and Amplification (IDA). This is a proposal for bootstrapping weaker models into stronger ones, developed at the intersection of capabilities and alignment research. Roughly speaking, the idea is to alternate between two stages: have the model think slowly to produce outputs better than its reflexes (amplification), then train on those high-quality outputs until the better judgment becomes reflex (distillation). IDA is analogous to the way one trains a graduate student to write papers - the first couple of papers take a dozen agonizing rewrites and editing passes, with the goal that the student can successfully distill the lessons from these laborious sessions and write essentially polished drafts on the first (or third) try.

Another reason I’ve become more excited about working in this space recently is that LLMs can now think about all the matrix multiplication for us. It’s been claimed that the primary value humans can now add to both capabilities and alignment research is “research taste,” the ability to select the right questions and introduce the most aesthetic formalizations. There’s nobody with better taste than mathematicians.3 Can I possibly compete with the legions of ambitious youngsters who grew up in this era?Mathematicians have a sort of deep learned helplessness about acting in the “Real World,” since we are trained to act within our narrow sandbox with very constrained imagination. When I was in grad school, for example, it occurred to me that if I wanted to maximize my contribution to mathematical progress, I would do better by being a personal assistant to my much more brilliant PhD advisor to free up his research time than by trying to prove theorems on my own. I know of someone else who was a full-time ghostwriter for other people’s papers who thus contributed more to the progress of their subfield than most of their peers.Learned helplessness sounds like a bad thing, but it’s actually like spending your whole life wearing heavy training weights. My observation is that mathematicians who successfully take these training weights off can become superstars. Acting with agency and imagination in the Real World is complicated and does require practice, but in the end this is a relatively easy and primitive mental skill compared to algebraic geometry. Math is the language in which the universe is written, and in many parts of the world the actual hardest part of acting effectively is being good at math. This is why many of the top AI capabilities and alignment researchers have deep math/TCS backgrounds.Let me also add that academic mathematicians have been partially responsible for (and complicit in) steering AI progress for a long time. Many of the most-cited benchmarks, including FrontierMath, FirstProof, and many others, were developed by mathematicians in collaboration with frontier labs, and their existence is likely partially responsible for the particular mathematical strength of current models. Math academia already has an ecosystem in place for designing evals to guide AI research; we should be using this ecosystem to advance alignment and the interests of humanity.4 OK, I’m convinced. What can I actually do to contribute?There are numerous existing ways to dip your toes in the water. Here are a few off the top of my head:Start a reading course. Is there anything mathematicians are better at than starting reading courses? Most PhD students and postdocs in mathematics are extremely worried about AI and their job prospects; it is essentially trivial to nudge them into starting reading groups and looking into alignment research, if only as a backup option in case mathematics goes kaboom.Contact your local AI safety group. Most research universities with computer science programs already have active AI safety initiatives and student groups; you can likely do a lot of good cheaply by contacting your local group and helping them reach mathematics departments, where they are relatively inactive.Look into short-term fellowships. There are plenty of reputable fellowship, visiting-researcher, and career-transition programs designed exactly to get outsiders up to speed on alignment research in a few months, without a career-ending commitment. Some such programs I (weakly) endorse are hosted by ARC, MATS, and Anthropic. Senior researchers can probably get even more mileage out of directly contacting organizations of interest.Apply for funding. There are already hundreds of millions in yearly https://harrywaterman.com/fieldmap/ available for safety research, and it is projected that this amount will shoot up by an https://harrywaterman.com/fieldmap/#context with the astronomical upcoming IPOs of certain AI labs. A surprising amount of such funding has already been captured by “one interested academic and their group.” Mathematicians may soon have an easier time being funded for mathematical research in AI safety via philanthropy than by the NSF.ResourcesHere are some places to start, whether you want to organize a reading group, try a research project, or take a few months to explore the field.Getting oriented and starting a reading grouphttps://harrywaterman.com/fieldmap/: Overview of the field, including organizations, research areas, headcounts, budgets, and funding flows.https://bluedot.org/courses: Online AI safety courses and project sprints, with material you can use to structure a reading group.https://www.alignmentforum.org/: Research posts and technical discussion across alignment agendas.https://www.arena.education/: Practical training and exercises for the empirical side of AI safety, including deep learning and mechanistic interpretability.Fellowships and research programshttps://www.matsprogram.org/: Mentored research across AI safety areas, including theory, with extension and residency pathways.https://alignment.anthropic.com/2025/anthropic-fellows-program-2026/: Four months of empirical AI safety research with funding and mentorship from Anthropic researchers.https://www.sparai.org/: A three-month, remote research program with flexible, part-time participation and mentorship in AI safety and policy.https://princint.ai/programs/fellowship/: Run by Principles of Intelligence; an approximately three-month interdisciplinary fellowship connecting researchers from fields such as mathematics, neuroscience, and philosophy with AI safety mentors.https://www.aialignmentfoundation.org/fellowship: An eight-week, remote, full-time alignment research fellowship.https://constellation.org/programs/astra: Full-time research with separate empirical AI safety and strategy and governance streams.

https://www.lesswrong.com/posts/PX4oTMHGa3XhFynJJ/ai-safety-a-short-faq-for-mathematicians-1#comments

https://www.lesswrong.com/posts/PX4oTMHGa3XhFynJJ/ai-safety-a-short-faq-for-mathematicians-1