Jeffrey M. Binder @jbinder.bsky.social · 05/10/2026What year was it—2023 I think—when it got up to like 70 degrees here in January? Could happen again 000
Jeffrey M. Binder @jbinder.bsky.social · 05/10/2026Let’s get drinks or something! I’ll likely be around 020
Jeffrey M. Binder @jbinder.bsky.social · 05/10/2026Looking at these replies, it’s striking how toxic Bluesky is getting, at least when the AI boogeyman comes up. Mindless friend/enemy battle-line drawing. 1100
Jeffrey M. Binder @jbinder.bsky.social · 02/10/2026Bandcamp Friday again, and this time I have a new batch of analogue/modular synth drones: circlesofproportion.bandcamp.com/album/nine-f...circlesofproportion.bandcamp.comNine Facilities Not Otherwise Specified, by Proportion9 track album 030
Jeffrey M. Binder @jbinder.bsky.social · 01/10/2026Right! A stochastic system with a fixed random seed still behaves like a stochastic system in many important ways. I feel like we lack a good way of expressing what we’re really concerned about and “deterministic” has stepped in as a substitute. 010
Jeffrey M. Binder @jbinder.bsky.social · 01/10/2026Kind of tired myself so I may have to pick this up tomorrow! 000
Jeffrey M. Binder @jbinder.bsky.social · 01/10/2026In any case, the term is spreading beyond the technical field and seems to be capturing an array of concerns that haven't really been sorted out clearly. 101
Jeffrey M. Binder @jbinder.bsky.social · 01/10/2026The result, it seems to me, is a general sense that deterministic systems are more trustworthy, which is half true. Some deterministic algorithms can be proven to get the right answer in all cases (e.g. long division), but others have varying levels of accuracy and some just don't work. 210
Jeffrey M. Binder @jbinder.bsky.social · 01/10/2026Nondeterministic systems require a different approach to testing because you can't just check the output against a single correct value. This has inspired broader concerns about how such systems could be made reliable. 100
Jeffrey M. Binder @jbinder.bsky.social · 01/10/2026In CS, a deterministic system is one that always behaves the same given the same input. Most current AI is nondeterministic in this sense because there's some degree of randomness in the output. 100
Jeffrey M. Binder @jbinder.bsky.social · 01/10/2026One trend I've noticed in industry is a desire to make systems "deterministic." This word seems to be straying from its computer-science definition into epistemic virtue territory, getting conflated with accuracy or certainty. Not exactly a reaction against AI, but a sort of footing readjustment. 150
Jeffrey M. Binder @jbinder.bsky.social · 01/10/2026Are you familiar with LangSmith? Not sure if it's exactly what you're after, but it stores structured traces of what agents are doing docs.langchain.com/langsmith/tr...docs.langchain.comTrace Claude Code applications - Docs by LangChain 110
Reposted by Jeffrey M. BinderJeffrey M. Binder @jbinder.bsky.social · 01/05/2026It’s Bandcamp Friday, a good day to point out that I released some weird synth music circlesofproportion.bandcamp.com/album/prelim...circlesofproportion.bandcamp.comPreliminal, by Circles of Proportion15 track album 011
Jeffrey M. Binder @jbinder.bsky.social · 06/07/2026I should have said $70k/yr IF they hit the jackpot on the job market 010
Jeffrey M. Binder @jbinder.bsky.social · 06/07/2026That'll show those humanities professors who make $70k/yr and are losing their jobs left and right /s 110
Reposted by Jeffrey M. BinderJeffrey M. Binder @jbinder.bsky.social · 12/06/2026New publication alert! I contributed a chapter to the book Dis/Trusting the Digital World in Imaginative Literature (@edinburghup.bsky.social). My chapter touches on Edgar Allan Poe’s “The Raven,” the Mechanical Turk hoax, and Charles Babbage's approach to establishing trust in machines. 051
Jeffrey M. Binder @jbinder.bsky.social · 12/06/2026New publication alert! I contributed a chapter to the book Dis/Trusting the Digital World in Imaginative Literature (@edinburghup.bsky.social). My chapter touches on Edgar Allan Poe’s “The Raven,” the Mechanical Turk hoax, and Charles Babbage's approach to establishing trust in machines. 051
Jeffrey M. Binder @jbinder.bsky.social · 06/06/2026I agree that referencing autocomplete can be helpful as a starting point, and I've used it before when teaching. But it only gets you so far in explaining what LLMs are doing unless you go into the details, like what “predict” means in this context, the role of training data, etc. 100
Jeffrey M. Binder @jbinder.bsky.social · 06/06/2026A problem with "LLMs are just autocomplete" as a means of explanation is that what autocomplete is doing—in terms of both objective and mechanism—is far from straightforward or widely understood. The analogy effectively amounts to explaining one black box by another black box. 2121
Jeffrey M. Binder @jbinder.bsky.social · 05/06/2026German word for the special anxiety one feels when interacting with the Phone app, where one wrong tap could result in accidentally calling someone 071
Jeffrey M. Binder @jbinder.bsky.social · 26/05/2026One of the first jazz musicians I got really into. 020
Reposted by Jeffrey M. BinderJosh Shepperd @joshshepperd.bsky.social · 26/05/2026RIP Sonny Rollins! My favorite musician and a genius. Time to spin St. Thomas. 312313
Jeffrey M. Binder @jbinder.bsky.social · 17/05/2026I spent several weeks doing basically nothing but checking the citations in my book. It wasn’t fun, but… 120
Jeffrey M. Binder @jbinder.bsky.social · 01/05/2026It’s Bandcamp Friday, a good day to point out that I released some weird synth music circlesofproportion.bandcamp.com/album/prelim...circlesofproportion.bandcamp.comPreliminal, by Circles of Proportion15 track album 011
Jeffrey M. Binder @jbinder.bsky.social · 21/04/2026Reminds me of this inverted digital clock I made a few years back: jeffreymbinder.net/apps/clock/jeffreymbinder.netThis is a clock where the hours move first and the minutes second. That is, each half-day (this is a US clock) is divided into 60 mours, each of which is divided into 12 hinutes. 100
Jeffrey M. Binder @jbinder.bsky.social · 16/02/2026I'm personally skeptical that hallucination can be solved in the current paradigm, but the Computer World article is basically misinformation, and people are jumping on it because it confirms their biases. 000
Jeffrey M. Binder @jbinder.bsky.social · 16/02/2026Plausibly, they might. I am not defending OpenAI's position. My point is that the Computer World article is misrepresenting its source, and we should raise our standards. 100
Jeffrey M. Binder @jbinder.bsky.social · 16/02/2026This seems to be making the rounds again, and the framing is highly misleading. The mathematical result only applies to “base” models that haven’t gone through RLHF. The study actually argues that hallucination is _not_ inevitable. www.computerworld.com/article/4059...computerworld.comOpenAI admits AI hallucinations are mathematically inevitable, not just engineering flawsIn a landmark study, OpenAI researchers reveal that large language models will always produce plausible but false outputs, even with perfect data, due to fundamental statistical and computational limi... 210
Jeffrey M. Binder @jbinder.bsky.social · 11/01/2026Clearly, this shows a man soldering his own finger to the soldering iron’s power supply 010
Reposted by Jeffrey M. BinderJeffrey M. Binder @jbinder.bsky.social · 26/12/2023This is your annual reminder that Sun Ra made a Christmas song www.youtube.com/watch?v=HEpF...youtube.comSun Ra presents The Qualities - It's Christmas Time (1960)A holiday tribute to Sun Ra fea, Doo Wop produced by Sun Ra on Saturn Records, 1960. Written by Alton Abraham and Sun Ra. 041
Jeffrey M. Binder @jbinder.bsky.social · 23/11/2025In 2021, I developed PromptArray, which lets you muck around with the internals of GPT models. I moved on because this method doesn't work with closed-source models like GPT-3, but GPT-OSS makes it possible again. Read if you miss the weirdness of the GPT-2 era! jeffreymbinder.net?p=480jeffreymbinder.netPromptArray, meet GPT-OSS! – Jeffrey M. Binder 0232
Jeffrey M. Binder @jbinder.bsky.social · 19/09/2025Still in New York, continuing to keep on. We should catch up some time! 010
Jeffrey M. Binder @jbinder.bsky.social · 19/09/2025There’s also Choco Leibniz, named after the inventor of choco calculus 1100
Jeffrey M. Binder @jbinder.bsky.social · 06/08/2025I made an online simulation of this method: jeffreymbinder.net?p=304jeffreymbinder.netArtificial Versifying – Jeffrey M. Binder 150
Jeffrey M. Binder @jbinder.bsky.social · 03/08/2025“Large language models like ChatGPT produce shallow, unoriginal ‘predictive text-y ideas’ and I worry that my students and others will increasingly believe that that’s okay—that there’s nothing better than that to aspire to.” 030
Reposted by Jeffrey M. BinderRyan Cordell @ryancordell.org · 22/01/2025I do not see how we’re going to have productive academic discussions of "AI" until we stop accepting the marketing that lumps all machine learning methods & technologies into one amorphous thing called "AI"—so we get minimalist & maximalist responses that echo that marketing—we don’t have to do this 1139190
Jeffrey M. Binder @jbinder.bsky.social · 23/11/2024Just thought of one. My mom saw a picture of a corpse-painted black metal band—I think it was Immortal—and said, "I remember bands like that. Like Kiss." 130
Jeffrey M. Binder @jbinder.bsky.social · 01/11/2024Is it possible for numbers to get too big? We’ll tell you about a shocking new development, after this. 060
Reposted by Jeffrey M. BinderMatthew Kirschenbaum @mkirschenbaum.bsky.social · 12/10/2024The Theories and Methodologies cluster on #criticalAI co-edited for PMLA by @ritaraley.bsky.social and myself is now online, open access. it consists of eight interventions by leading voices, as well as our framing essay “AI and the University as a Service.” www.cambridge.org/core/journal...cambridge.orgFirstView | PMLA | Cambridge CorePMLA - Brent Hayes Edwards 89355
Jeffrey M. Binder @jbinder.bsky.social · 30/09/2024Today I learned the word "ultracrepidarianism" 010
Jeffrey M. Binder @jbinder.bsky.social · 25/09/2024Not sure about Llama3, but have you looked at the structured output modes offered by some LLMs? python.langchain.com/docs/how_to/...python.langchain.comStructured Output | 🦜️🔗 LangChainIt is often crucial to have LLMs return structured output. This is because oftentimes the outputs of the LLMs are used in downstream applications, where specific arguments are required. Having the LLM... 160
Jeffrey M. Binder @jbinder.bsky.social · 24/09/2024My book made it to the Chicago Manual of Style! I don't know how I'm going to top this. 050
Jeffrey M. Binder @jbinder.bsky.social · 17/09/2024If you see this, quote skeet with a famous landmark you’ve seen. 000
Jeffrey M. Binder @jbinder.bsky.social · 08/09/2024Thinking of the time when I bought two CDs from a stand at the WFMU record fair—"Momentum" by Monolake and "Jonathan Sings!" by Jonathan Richman—and the seller seemed to have trouble comprehending that the same person could want both of these albums. 010
Jeffrey M. Binder @jbinder.bsky.social · 07/09/2024This post goes really well with the Wittgenstein I was just reading 010