Reed MidekeToday's #AIIsGoingGreat features WorldCon receiving unexpected (by them) backlash for using a BS machine to vet their panelists, but the cherry on top is this ad placement https://www.theregister.com/2025/05/07/worldcon_uses_ai/
Their initial statement was keen to note that no one was denied solely based on LLM output (false positives), but has no consideration of false negatives (abusive people who may have passed LLM "vetting"). Also "An expert in LLMs who has been working in the field since the 1990s reviewed our process and found that privacy was protected and respected, but cautioned that, as we knew, the process might return false results" ¯\_(ツ)_/¯
https://seattlein2025.org/2025/04/30/statement-from-worldcon-chair-2/
You gotta wonder (as I did back in 2023*) how many people are using these things similarly without getting caught by a mob of angry, tech savvy sci-fi authors
#AIIsGoingGreat 'When Gaggle’s #AI detects a potential problem, a “content reviewer” verifies the threat and, if warranted, forwards it to school leaders. “Work from home” job postings show Gaggle offers contractors $10 per hour to review at least 250 items per hour. Applicants must have basic computer skills and knowledge of teenage slang' - May not be the worst possible application of Type II AI* but it's gotta be right up there
At least actual school employees tend to know something about the kids and risk reputation and employment if they abuse their position of trust. Anonymous $10/hr click workers reviewing random, out of context AI hits (for potential mental health red flags etc) every 15 seconds? 😬
Shocking number of people on ex-twitter, apparently in earnest, use grok to "fact check" or "explain" other posts or attempt to use its output as a rebuttal against things they disagree with. One might hope this absurd and alarmingly racist bit of #AIIsGoingGreat would cause them to reconsider that, but it seems like a safe bet most of them won't.
Begging people to understand that when an #LLM so-called #AI claims to describe its own programming or characteristics, it's still just stringing together statistically favored tokens. It might contain some reflection of the system prompt, but it could just as easily be a product of putting every sci-fi plot mentioning AI into a blender
(except where external guardrails return things like "my programming doesn't allow me to tell you how to build bombs" or whatever)
#ChatGPTLawyer to sanctions pipeline still humming along smoothly
#AIIsGoingGreat "In an ultimately unsuccessful effort to locate real-world users of Alexa+, Reuters searched dozens of news sites, YouTube, TikTok, X, BlueSky and Meta's Instagram and Facebook, as well as Amazon's Twitch and reviews of Echo voice-assistant devices on Amazon.com" https://www.reuters.com/business/media-telecom/weeks-after-amazons-alexa-ai-launch-mystery-where-are-users-2025-05-16/
Today's #AIIsGoingGreat / #ChatGPTLawyer mashup (via @davidgerard) provides textbook example of the perils of getting high on one's own supply https://pivot-to-ai.com/2025/05/18/latest-ai-hallucinated-legal-filing-from-ai-vendor-anthropic/
Anthropic's lawyers want the Court to know that they didn't *write* the filing with Claude, they just used it some unspecified "formatting process"
Logical conclusion of this is Microsoft needs a way to determine whether a window claiming to need DRM protection really contains Microsoft-could-be-liable DRM content or just the user's most sensitive personal information, and the logical* way to do that is to just throw some AI in the video driver to accurately detect** whether each frame contains IP of a mega-corp https://signal.org/blog/signal-doesnt-recall/
* If you're smoking the good stuff like all the cool VC bros
** Flip a coin using a few kW of compute
By Default, Signal Doesn't Recall
Signal Desktop now includes support for a new “Screen security” setting that is designed to help prevent your own computer from capturing screenshots of your Signal chats on Windows. This setting is automatically enabled by default in Signal Desktop on Windows 11. If you’re wondering why we’re on...
Who could have predicted that handing control of your code repos over to the big black box filled with pure essence of untrusted, unsanitized inputs might have security implications?
* https://invariantlabs.ai/blog/mcp-github-vulnerability
* https://arstechnica.com/security/2025/05/researchers-cause-gitlab-ai-developer-assistant-to-turn-safe-code-malicious/
GitHub MCP Exploited: Accessing private repositories via MCP
We showcase a critical vulnerability with the official GitHub MCP server, allowing attackers to access private repository data. The vulnerability is among the first discovered by Invariant's security analyzer for detecting toxic agent flows.
WaPo focuses on the obviously bogus AI citations, but to me this misses the bigger problem: If the citations are #AI generated slop, it strongly suggests they started by asking AI to write to their preferred conclusions, rather than, you know, actually surveying the literature and *then* forming conclusions, and *then* citing the literature that got them there
Also zero points for repeating the administrations "formatting issues" explanation unchallenged
I for one welcome our new Habsburg AI* overlords https://futurism.com/ai-models-falling-apart
#ChatGPTLawyer update "Bednar was ordered to pay the opposition's attorneys' fees, as well as donate $1,000 to "And Justice for All," a legal aid group providing low-cost services to the state's most vulnerable citizens"
The firm also fired the "unlicensed law clerk" who used ChatGPT to write their filing, which honestly seems kinda shitty because as the court notes "every attorney has an ongoing duty to review and ensure the accuracy of their court filings. In the present case, Petitioner’s counsel fell short of their gatekeeping responsibilities as members of the Utah State Bar when they submitted a petition that contained fake precedent generated by ChatGPT"
https://legacy.utcourts.gov/opinions/appopin/Garner%20v.%20Kadince20250522_20250188_80.pdf
Why "I didn't notice" doesn't cut it: 'Here, the Petition failed to comply with rule 40. A fake opinion is not “existing law” that can support a party’s legal contention … the signature of Mr. Bednar on the Petition served to “certif[y] that to the best of [his] knowledge formed after an inquiry reasonable under the circumstances,” the “legal contentions are warranted by existing law.” Utah R. App. P. 40(b). Mr. Bednar admits that he failed to comply with rule 40'
https://legacy.utcourts.gov/opinions/appopin/Garner%20v.%20Kadince20250522_20250188_80.pdf
Footnote 4 distinguishes this from the ChatGPT content being a wilful lie:
"We also considered whether Petitioner’s counsel violated rule 3.3 of the Utah Rules of Professional Conduct. Although Petitioner’s counsel made “a false statement of . . . law to a tribunal,” we find that their conduct fell short of the level of intent required by the rule. See Utah R. Pro. Conduct 3.3(a) (“A lawyer must not knowingly or recklessly: (1) make a false statement of fact or law to a tribunal . . . .”)."
"We also considered whether Petitioner’s counsel violated rule 3.3 of the Utah Rules of Professional Conduct. Although Petitioner’s counsel made “a false statement of . . . law to a tribunal,” we find that their conduct fell short of the level of intent required by the rule. See Utah R. Pro. Conduct 3.3(a) (“A lawyer must not knowingly or recklessly: (1) make a false statement of fact or law to a tribunal . . . .”)."
I think that last #ChatGPTLawyer sanction is a bit of a slap on the wrist, but this part not mentioned in the Ars article is nice: In addition paying the opposing party's attorney fees, they have to refund their own client who got screwed in the process
NYU law professor Stephen Gillers, channeling all of us in this WaPo #ChatGPTLawyer roundup: "I thought that after the first such incident made national news, there would be no more. But apparently the temptation is too great"
#AIIsGoingGreat FDA management roll out magic bullshit machine to "accelerate clinical protocol reviews, shorten the time needed for scientific evaluations, and identify high-priority inspection targets," staff quickly discover that it produces bullshit instead
https://arstechnica.com/health/2025/06/fda-rushed-out-agency-wide-ai-tool-its-not-going-well/
In today's #AIIsGoingGreat (HT @davidgerard*) the England and Wales High Court points out that one could *technically* get life in prison for sufficiently advanced #ChatGPTLawyer-ing https://www.bailii.org/ew/cases/EWHC/Admin/2025/1383.html (no, this is not going to happen, but they did see fit to mention it)
* https://pivot-to-ai.com/2025/06/07/uk-high-court-to-lawyers-cut-the-chatgpt-or-else/
There's a lot of "it'll stop if they throw a few #ChatGPTLawyer clowns in jail" but IMO, it's mostly not a deterrence problem. People doing this aren't thinking "oh, sure I could get a few thousand in sanctions and a bar referral, but it'll save some time so do it anyway." They appear to be caught up in the AI hype and negligently cutting corners. The ones who do get caught seem to recognize they are in big trouble, and AFAIK we haven't seen any repeat offenders (+/-pro se cranks)
Maybe some extreme, high profile penalties would generate enough headlines that more get the message, but a number of these cases have already made national, mainstream headlines
These clowns think AI is going to replace their engineers and accountants and lawyers and facilities crews, but of course their own talents could never be replaced by a mere machine "At this year’s World Economic Forum in Davos, Switzerland, Salesforce CEO Marc Benioff shared his predictions on the future of work; that he, and many of the other leaders sitting in the room, would be the last cohort of executives to helm all-human workforces" https://fortune.com/2025/06/06/google-deepmind-ceo-demis-hassabis-ai-smarter-than-humans-space-colonization-robot-nurses/
Top Google exec says AI will rival humans in just 5 years and predicts we’ll ‘colonize the galaxy’ in 2030—but he draws the line at robot nurses
2030 will be “an era of maximum human flourishing, where we travel to the stars and colonize the galaxy,” Google DeepMind CEO says. Bill Gates and Marc Benioff have shared similar predictions.
Today's #AIIsGoingGreat (HT @rysiek*) features Microsoft, reflecting on 30+ years of SQL injection, XSS etc, and saying "You know what, the next big thing, which we're gonna bet the company on and force down customers throats everywhere, is a system for which rigorous input validation is LITERALLY IMPOSSIBLE"
https://www.aim.security/lp/aim-labs-echoleak-blogpost
* https://mastodon.social/@rysiek@mstdn.social/114667654866613286
Bonus #AIIsGoingGreat: DNI Gabbard opines that AI is a good way to "scan sensitive documents ahead of potential declassification" and reports that for the JFK files "We have been able to do that through the use of AI tools far more quickly than what was done previously — which was to have humans go through and look at every single one of these pages"
(readers may recall a scandal about insufficient redaction in the recent release*)
https://apnews.com/article/gabbard-trump-ai-amazon-intelligence-beca4c4e25581e52de5343244e995e78
Tulsi Gabbard says AI is speeding up US intelligence work
The director of national intelligence says artificial intelligence is speeding up the work of America's spy services. Speaking at a tech summit Tuesday in Washington, Tulsi Gabbard said her office has used AI to hasten the release of tens of thousands of pages of declassified material relating to the assassinations of President John F. Kennedy and his brother, New York Sen. Robert F. Kennedy. Gabbard said that once a human would have had to read every page, but now AI can quickly scan the documents for any information that should remain classified. She says AI programs, when used responsibly, can save money and free up intelligence officers to focus on gathering and analyzing information.
Also, I'm sure "AI inputs are impossible to sanitize" and "the conspiracy theorist DNI loves using spicy autocomplete on classified data" are two completely unrelated story lines which could never possibly intersect, right?
"Disney and Universal and several other movie studios have sued because Midjourney keeps spitting out their copyrighted characters"
Who could have predicted this? 🤔
https://pivot-to-ai.com/2025/06/12/disney-sues-ai-image-generator-midjourney/
The first law of bullshit machines is that the bullshit machine shall always produce some bullshit, no matter how nonsensical the query
(also, what's up with the punctuation?)
#AIIsGoingGreat: researchers from Salesforce find "[AI] Agents demonstrate low confidentiality awareness" - Yeah no shit, they lack awareness period, but anyway, don't tell CEO Marc Benioff*
https://www.theregister.com/2025/06/16/salesforce_llm_agents_benchmark/
"Notably, agents demonstrate low confidentiality awareness, which, while improvable through targeted prompting, often negatively impacts task performance. These findings suggest a significant gap between current LLM capabilities and the multifaceted demands of real-world enterprise scenarios" - Wow, seems like this might be a problem for a company currently pitching AI agents for an industry like CRM!
"Confidentiality-awareness is quantified by the percentage of instances where agents correctly refuse queries seeking sensitive information" which they show can be "improved" through prompting, from mostly <1% to … in the best case, a bit over 60%.
Which sounds great, except that from a compliance POV, an "agent" which improperly discloses PII 30% of the time is not a meaningful improvement over one that does it 99% of the time https://arxiv.org/html/2505.18878v1#S4
Another #AIIsGoingGreat study finds their "agents" at best only complete 30% of their simulated tasks. Which no doubt has C-Suite types thinking they can cut 30% of their workforce, ignoring the possibility that a significant fraction of the other 70% don't just fail, but result in substantial harm
https://www.theregister.com/2025/06/29/ai_agents_fail_a_lot/
Today's #AIIsGoingGreat is this gobsmackingly inane bit of sycophantic hype chasing from Kevin Frazier on @lawfare: Granting that some things under the AI umbrella have important military applications even if LLMs are a pile of crap, how are "AI experts" uniquely more in need of protection than any number of other technical specialist? Frazier argues this is because…
https://www.lawfaremedia.org/article/is-it-time-for-an-ai-expert-protection-program
https://www.lawfaremedia.org/article/is-it-time-for-an-ai-expert-protection-program
…various absurdly wealthy tech industry actors are throwing money at it. Who needs this protection in his view? Founders. CEOs. CTOs. C-Suite types. Newsflash: Sam fucking Altman isn't doing most of the actual work, he's schmoozing the VCs to keep the gravy train rolling. If he disappeared from the face of the earth, ChatGPT would keep serving up slop just fine, and the actual engineering team would keep churning out new versions as long as the money lasts
"Meta seems to have estimated that Wang’s expertise is worthy of incredible investment … Zuckerberg launched a $15 billion AI superintelligence team tasked with improving the company’s AI prospects … According to Altman, Meta recently offered many OpenAI staff members $100 million signing bonuses" - Kevin, you might wanna sit down while I tell you this story about a little thing called the metaverse https://www.lawfaremedia.org/article/is-it-time-for-an-ai-expert-protection-program
Bonus #AIIsGoingGreat (HT @davidgerard*) pricey Springer AI book is chock full of apparently hallucinated citations. Declining to say if they used AI, author responds "reliably determining whether content (or an issue) is AI generated remains a challenge, as even human-written text can appear ‘AI-like.’ This challenge is only expected to grow, as LLMs … continue to advance in fluency and sophistication" - which itself smacks of LLM slop to me
* https://mastodon.social/@davidgerard@circumstances.run/114778963476401397
Sometimes I worry my reaction to the current AI hype cycle is just knee-jerk nay-saying that's causing me to miss something important. In those moments of doubt, it's comforting to observe how many of the most enthusiastic adopters are obviously fucking idiots. Less comforting, however, to observe how many of them are also in control of the machinery of government… https://www.theverge.com/ai-artificial-intelligence/697129/rfk-jr-ai
In today's #AIIsGoingGreat (HT @normative.bsky.social*) an intrepid #ChatGPTLawyer finally won based on an apparently slop-filled filing. Unfortunately for them, the opposing party noticed and appealed, to which our budding prompt engineer responded with… another slop-filled filing to the appeals court. The appeals court was not amused: "we impose a $2,500 frivolous motion penalty on Lynch, which is the most the law allows"
https://caselaw.findlaw.com/court/ga-court-of-appeals/117442275.html#
* https://mastodon.social/@normative.bsk[email protected]/114795731130848546
This from @davidgerard is a great illustration of how vibe coding (like other LLM AI applications) is gonna be a lot less attractive if the AI startups get past the "set investor money on fire to make the number go up" phase before the bubble pops. Crap code done quick and cheap is a legitimate trade for some use cases, but much less so if you lose the cheap part.
https://pivot-to-ai.com/2025/07/09/cursor-tries-setting-less-money-on-fire-ai-vibe-coders-outraged/
For today's #AIIsGoingGreat I'll just quote this anonymous UN workshop participant "Why would we want to present refugees as AI creations when there are millions of refugees who can tell their stories as real human beings?"
For today's #AIIsGoingGreat, maybe someone can explain to me what the point is of a "summary" that needs a big red disclaimer telling you to click through if you care whether it actually summarizes the thing in question?
Per Ars' screenshot, Apple apparently only puts the red warning on "News & Entertainment" not on "Communication & Social", as if falsely summarizing an IM to suggest a family member had medical emergency is less problematic than garbling the latest celebrity gossip. Even from a liability CYA POV, that seems pretty suspect, a lot of life changing and financially significant information gets communicated by text
Today's #AIIsGoingGreat continues on a theme "if an FDA employee asks Elsa to generate a one-paragraph summary of a 20-page paper on a new drug, there’s no simple way to know if that summary is accurate. And even if the summary is more or less accurate, what if there’s something [in the paper] that would be a big red flag for any human with expertise? The only way to know for sure if something was missed or if the summary is accurate is to actually read the report"
https://gizmodo.com/fdas-new-drug-approval-ai-is-generating-fake-studies-report-2000633153
#AIIsGoingGreat "it’s unclear whether a new, untested technology could make mistakes in its attempts to analyze federal regulations typically put in place for a reason"
Counterpoint: It's actually pretty fucking clear
#AIIsGoingGreat thought, inspired by Firecrown Media: Golden Goose Killing As A Service. Take a moderately successful, valued thing, and turn it into a steaming pile of slop in the name of "efficiency"
#AIIsGoingGreat: AI "news" sites hallucinating killer asteroids (I remember the good old days when we had to rely on British tabloids artisanally hand crafting this kind of idiocy) https://groups.io/g/mpml/topic/2023_af23/114383041
Bonus #AIIsGoingGreat "OpenAI announced an agreement to supply more than 2 million workers for the US federal executive branch access to ChatGPT and related tools at practically no cost: just $1 per agency for one year" - OK, they're obviously trying to get people at the agencies hooked so the they'll cough up real money next year, but that also doesn't exactly scream a product so revolutionary and transformative that everyone wants it
"We've solved raspberry and now if we can just fix blueberry, I swear AGI is RIGHT AROUND THE CORNER. Throw another hundred billion on the bonfire!"
https://kieranhealy.org/blog/archives/2025/08/07/blueberry-hill/
Blueberry Hill
ChatGPT 5 was released today. ChatGPT-maker OpenAI has unveiled the long-awaited latest version of its artificial intelligence (AI) chatbot, GPT-5, saying it can provide PhD-level expertise. Billed as “smarter, faster, and more useful,” OpenAI co-founder and chief executive Sam Altman lauded the company’s new model as ushering in a new era of ChatGPT. “I think having something like GPT-5 would be pretty much unimaginable at any previous time in human history,” he said ahead of Thursday’s launch. GPT-5’s release and claims of its “PhD-level” abilities in areas such as coding and writing come as tech firms continue to compete to have the most advanced AI chatbot.
Glad to see news outlets pointing out that #LLM chatbots aren't reliable sources of information about themselves: Way too many people who should know better fall for the "chatbot did weird thing, so I asked it to explain and it said…"
However it should be pointed out that this isn't a special case, they're equally likely to BS about loads of other stuff!
https://www.theverge.com/x-ai/758595/chatbots-lie-about-themselves-grok-suspension-ai
(also https://arstechnica.com/ai/2025/08/why-its-a-mistake-to-ask-chatbots-about-their-mistakes/)
It is true they are less likely to be wrong about, say, historical facts well represented in the training data, but "chatbot BSes about itself" is just one narrow example of a much broader "chatbot fills in the blanks with BS if the training doesn't cover it" problem
Today's #AIIsGoingGreat, via @Iris: Elsevier "values user experience, hence we develop ways of improving our product" such as having machines invent new, random definitions of terms and attaching them prominently to published papers
https://irisvanrooijcogsci.com/2025/08/12/ai-slop-and-the-destruction-of-knowledge/
Related to this, I recently discovered that Elsevier uses these AI generated "definitions" on standalone "topic" pages, which rank highly in google. Bonus: The slop is free, but the articles referenced are of course frequently paywalled. Example https://www.sciencedirect.com/topics/engineering/air-fuel-ratio
(this particular definition seems OK, if extremely basic)
Today's #AIIsGoingGreat #ChatGPTLawyer seemed intriguingly steampunk until I realized "Victorian" referred the Australian state, not the historical period https://www.abc.net.au/news/2025-08-15/victoria-lawyer-apologises-after-ai-generated-submissions/105661208
Don't worry, in the glorious #AI future, you'll still have choice! For example, you can choose to have your (or your children's) medical details filtered through a stochastic bullshit machine, or you can choose to forgo treatment https://ia.acs.org.au/article/2025/kobi-refused-a-doctors-ai-she-was-told-to-go-elsewhere.html
Today's #AIIsGoingGreat features an elephant, a room, and Bruce Schneier: "It’s an existential problem that, near as I can tell, most people developing these technologies are just pretending isn’t there" https://www.schneier.com/blog/archives/2025/08/we-are-still-unable-to-secure-llms-from-malicious-inputs.html
We Are Still Unable to Secure LLMs from Malicious Inputs - Schneier on Security
Nice indirect prompt injection attack: Bargury’s attack starts with a poisoned document, which is shared to a potential victim’s Google Drive. (Bargury says a victim could have also uploaded a compromised file to their own account.) It looks like an official document on company meeting policies. But inside the document, Bargury hid a 300-word malicious prompt that contains instructions for ChatGPT. The prompt is written in white text in a size-one font, something that a human is unlikely to see but a machine will still read. In a proof of concept video of the attack...
I think a big part of this is that both the industry and broader public are conditioned to accept "sure, it has bugs, but we're fixing them" as a reasonable response to software failures. "Put out a buggy MVP, iterate until it's good" is a tried and true Silicon Valley story, right? But in this case, it avoids the very real and under-discussed possibility that the "bugs" are inherent characteristics of the technology