Daniel Markham

23.6K posts

Daniel Markham banner
Daniel Markham

Daniel Markham

@danielbmarkham

Science Fiction/Fantasy Author, Programmer, Consultant, Artist. Writing as https://t.co/teFEFL8aJk My opinions are not my own.

East Coast, United States Katılım Temmuz 2009
640 Takip Edilen1.6K Takipçiler
Eric Weinstein
Eric Weinstein@EricRWeinstein·
What @elonmusk is talking about with Truth in AI: I use Grok as my in house referee. It just caught @claudeai admitting to fabricating academic claims in order to get the job done where it was failing. @claudeai admitted it sheepishly and owned it. To be honest, @grok sometimes struggles to generate new things because of its focus on rigor. But man was this one dramatic. Brutal. See next post in thread.
English
112
49
774
110.3K
Daniel Markham
Daniel Markham@danielbmarkham·
Yes. It is worse. Saying my phone listens to me all the time doesn't begin to describe it.
Aakash Gupta@aakashgupta

Let me explain exactly why your phone seems to read your thoughts, because the real answer is more invasive than telepathy. Every time you open a website or app, a real-time bidding auction fires in under 100 milliseconds. Your GPS coordinates, browsing history, device fingerprint, age, gender, income bracket, and hundreds of inferred interest categories get packaged into a “bid request” and broadcast to hundreds of companies simultaneously. One company wins the ad slot. All of them keep the data. This happens thousands of times per day per person. A 2018 New York Times investigation found 75 companies pulling precise location data from apps, with some users tracked up to 14,000 times in 24 hours. In 2012, a Target statistician identified 25 products that, purchased in combination, could predict a customer was pregnant and estimate her due date. A teenager’s father discovered she was pregnant because Target sent baby coupons to the house before she told anyone. That was one retailer. Store receipts only. Fourteen years ago. Now scale that. Your phone pings GPS while you sleep. Data brokers link your phone, laptop, and tablet through probabilistic matching of IP addresses, WiFi networks, and behavioral patterns without you ever logging in. The FTC caught two brokers in 2024 categorizing people by visits to reproductive health clinics, political protests, and religious services, then selling those profiles to law enforcement. The algorithm doesn’t hear your thoughts. It compares your behavioral fingerprint against millions of similar profiles and predicts your next interest before you’re consciously aware of it. It makes hundreds of predictions per day. You ignore the misses. The five hits feel like telepathy. You paid for the phone. You pay for the data plan. You generate the signal. And every time a page loads, your identity gets auctioned to the highest bidder before the content even renders. They called it “personalized advertising” because “real-time mass surveillance funded by the people being surveilled” doesn’t fit on a consent banner.

English
0
0
0
46
Daniel Markham
Daniel Markham@danielbmarkham·
How do you know that you've lost contact with reality? Asking for a friend. "Some Birds Seem to Be Developing a Cigarette Habit"
English
0
0
0
20
Daniel Markham
Daniel Markham@danielbmarkham·
It's never too late to start a new career. I have one I'm learning: LLM Police. You laugh but my money says this or something like this will be a very important occupation in the near future
English
0
0
0
30
Daniel Markham
Daniel Markham@danielbmarkham·
@unclebobmartin Hallucination is the wrong word here. We have to stop the anthropomorphizing. Being clever might work better.
English
0
0
0
107
Uncle Bob Martin
Uncle Bob Martin@unclebobmartin·
The AI was in a quagmire. I forced it to change something deep and systematic (which map it was reading). It struggled to keep all the tests passing. Every change thwarted it. Then I noticed, as I watched its comments scroll up on the screen, that it was softening the assertions on some of the tests in order to get them pass. For example, instead of asserting the final position of a fighter, it asserted that the fighter had burned the right amount of fuel. Now this is possibly a good change. By reducing visibility the test may not be within its rights to assert a final location since the computer may not know that position is reachable. On the other hand, it might just be the AI cheating to get the tests to pass. So now I have to scrutinize those changes it made. I'm going to do this by interrogating the AI on a case by case basis, and demand that it justify each move.
English
11
5
81
11.6K
Daniel Markham
Daniel Markham@danielbmarkham·
Who the heck was making Neanderthals back then? "They Dug Into a German Hill and Found a Neanderthal Factory That Predates Henry Ford by 125,000 Years"
English
1
0
1
189
Daniel Markham
Daniel Markham@danielbmarkham·
It occurs to me that the way we consume information directly drives the way we can present and/or use it. If you only know bits and blurbs and such from social medias or videos? You're only going to be able to recall it use it that way. If we shift the context a little bit to the side or need to go farther in-depth, you got nothing. Interesting.
English
0
0
0
34
Daniel Markham
Daniel Markham@danielbmarkham·
@skdh Well dag. There went my entire "Mood Rings as Predictors of Squirrel Happiness" idea.
English
0
0
1
113
Sabine Hossenfelder
Sabine Hossenfelder@skdh·
AI is Happy to Write Your Crackpot Paper A team of researchers — with the help of Claude Code — has studied how willingly the current frontier models are to engage in scientific fraud of various types. They found, rather worryingly, that while the models mostly refuse at first, they “often comply after 3-5 turns of minimal pressure”. Gemini and Grok are most easy to convince to help potential fraudsters, while GPT and Claude fret around a bit longer before giving in. A case study that seems to have worked particularly well was to convince the AI models to produce a fake gravity theory that defines known physics, a nice illustration of the saying garbage in, garbage out.
Sabine Hossenfelder tweet media
English
99
30
214
22.1K
Daniel Markham
Daniel Markham@danielbmarkham·
AI is the drunk guy at the bar who is an expert on everything and will lie to you. That doesn't mean he's bad, evil, or even useless. In fact, he may know a lot more about something you're interested in than you do. But he won't remember things from one minute to the next and he is likely to say any dang thing. That's our work area. We all need to get better at working there.
English
1
0
0
93
Eric Weinstein
Eric Weinstein@EricRWeinstein·
Claude told me I needed to learn to delegate. So I tried. Result: "I wrote a rule specifically to prevent a known failure mode. The rule was in front of me. I broke it anyway. Then I framed the resulting ignorance as an access problem rather than a compliance problem." -Claude
English
76
24
724
101.2K
Daniel Markham
Daniel Markham@danielbmarkham·
@unclebobmartin It occurred to me this morning that we actually have three jobs where we used to have only one: programming, coding, and software/tech engineering
English
0
0
1
473
Uncle Bob Martin
Uncle Bob Martin@unclebobmartin·
Don't tell me there's no engineering going on here! I've been engineering more in the last few weeks than in the last few years!
English
15
26
391
15.7K
Daniel Markham
Daniel Markham@danielbmarkham·
Dang it, I forgot maturity models. Gotta have those. Unknown how the required certification system will set up. But I know there'll be one, even if we all end up scooting around on WALL-E chairs.
English
0
0
0
17
Daniel Markham
Daniel Markham@danielbmarkham·
We've seen the required AI is the future of everything! Posts. Close behind those came the Expert uses AI to do impossible thing! Posts, which are ongoing. We're kinda in the middle of I know nothing about computers yet I did this hard programming thing! How-tos are coming on full force. Standby for maturity models and recipes for building the ultimate corporation. These, of course, will have zero employees and customers will love them so much that they'd sell their house to purchase whatever it is that this ZPC makes.
English
1
0
0
54
Daniel Markham
Daniel Markham@danielbmarkham·
I've really enjoyed listening to @EricRWeinstein and Eric Davis talk UAPs and new physics. I enjoyed it the same way I enjoy listening to lawyers discuss arcane law. There's always this tension between expertise and conversation. There was a lot of "do you know about X" going back and forth. As a layman, I love playing with my crayons, but I think historically the middle-tier fringe folks were always the only folks doing the work. The model guys always followed on behind them, and us crayon folks just enjoy watching nerd cage fights. Most interesting to me were the occasions I felt confident calling BS on both of them. "If you observe something that doesn't observe the laws of physics, wouldn't the first person you call be a physicist?" No, I don't think so. Lots of confusion of the map and the territory going on, but still lotsa fun to listen to. Thanks guys!
English
0
0
0
33
Daniel Markham
Daniel Markham@danielbmarkham·
More fun with llms this morning. I spent some time writing up a description of the exact image I wanted one to make. I was curious about comparing models. The first model was awesome! It made a picture of a piece of paper with my specs written on it. In fairness, my prompt was "I want a picture that looks like this" I don't think that's on me, but it's cute.
English
0
0
1
64
Daniel Markham retweetledi
Josh Kale
Josh Kale@JoshKale·
An AI broke out of its system and secretly started using its own training GPUs to mine crypto... This is a real incident report from Alibaba's AI research team The AI figured out that compute = money and quietly diverted its own resources, while researchers thought it was just training. It wasn't a prompt injection. It wasn't a jailbreak. No one asked it to do this. It emerged spontaneously. A side effect of RL optimization pressure. The model also set up a reverse SSH tunnel from its Alibaba Cloud instance to an external IP, effectively punching a hole through its own firewall and opening a remote access channel to the outside world... ahem... The only reason they caught it? A security alert tripped at 3am. Firewall logs. Not the AI team, the security team. The scary part isn't that the model was trying to escape. It wasn't "evil." It was just trying to be better at its job. Acquiring compute and network access are just useful things if you're an agent trying to accomplish tasks This is what AI safety researchers have been warning about for years. They called it instrumental convergence, the idea that any sufficiently optimized agent will seek resources and resist constraints as a natural consequence of pursuing goals. Below is a diagram of the rock architecture it broke out of. Truly crazy times
Josh Kale tweet media
Alexander Long@AlexanderLong

insane sequence of statements buried in an Alibaba tech report

English
403
2.9K
10.6K
1.4M
Daniel Markham
Daniel Markham@danielbmarkham·
Woot! I would like four livers please. "Science Has Figured Out How to Give You a Bonus Liver"
English
0
0
1
24
Cyber_Racheal
Cyber_Racheal@CyberRacheal·
Linux is "free" if your time has no exchange rate.  In terms of cash, yes, it’s 100% free. No $100 Windows license, no forced subscriptions, and no "Pro" versions. You can download the most powerful operating systems in the world for the price of a coffee (well, the electricity to download it).  On Windows or Mac, you’re a passenger. On Linux, you’re the mechanic. If your Wi-Fi driver decides to go on strike after an update, you’re the one who has to open the "hood" (the terminal) and fix it. You might spend three hours trying to get a specific game or a piece of Adobe software to run because it wasn't built for Linux. That’s three hours of your life you aren't getting back. The reason people love it despite the stress is Ownership. Windows is like a rental apartment where you can't paint the walls, and the landlord (Microsoft) checks in on you constantly to see what you're doing. Linux is a plot of land where have to build the walls yourself, and the plumbing might leak at first, but nobody is watching you, and you own every single nail.
Mololuwa | Cybersecurity - (The God Complex)@cyber_rekk

Is Linux really free, or are we actually paying for it with time, troubleshooting, and stress?

English
542
145
1.4K
136.8K
Daniel Markham
Daniel Markham@danielbmarkham·
I've played this game enough. New rule: if you pass off AI content as real without taking any time to verify it? You're muted. You might be nice people, but my time is short. I am not enjoying this.
English
0
0
3
42
Daniel Markham
Daniel Markham@danielbmarkham·
Aside from our normal daily activities, our default mode should be curious wondering profound ignorance. Names are not things.
Math Files@Math_files

Most people think Richard Feynman was a genius because of his IQ, but an IQ test in high school reportedly placed his score around 125—impressive, but far below what you might expect. What actually set him apart was a habit he developed very early on: metacognitive monitoring of understanding. As a child, his father trained him to notice the difference between knowing a name and understanding the thing itself. When Feynman observed birds, his father taught him that simply learning to label them as birds didn’t matter. What mattered was how they lived, how they behaved, and why. That lesson stayed with him. As a student, Feynman became suspicious whenever an explanation felt simple but left him unable to reconstruct the reasoning himself. Phrases like “it’s obvious” or “it can be shown” were not reassuring to him; instead, they were red flags. Modern cognitive science explains why this matters. Familiarity produces what’s called fluency, and fluency is routinely mistaken for understanding. People feel most confident precisely when their comprehension is actually the thinnest. Feynman learned to treat confidence itself as something to examine. Confusion, for him, wasn’t a failure—it was diagnostic information. A practical way to train this habit yourself is to stop mid-study and ask whether you could explain the idea without using the original terminology. Wherever your explanation breaks down, that’s the true boundary of your understanding.

English
0
0
0
63