Wulfy—Speaker to the machines
Jan Wildeboer 😷
As someone who works in marketing, this is one of the cheapest tricks to sell something: "Our product is so good that it is dangerous and we cannot release it to the world". This translates to "Give us a lot of money for limited, exclusive access, sign this NDA (Non-Disclosure Agreement) so no one finds out it's not that good until we make enough revenue and can claim we made a less dangerous version for all, but the better version with more risks is established as high-price product." ;)
And of course, in "AI" land, use the established "you are prompting it wrong" when customers complain that they don't see the benefits of paying a lot for something that doesn't deliver ;)
C.Suthorn 
@[email protected]
Ich habe ein LLM gebaut, da wird der falsche Prompt eingegeben und die Antwort ist der richtige Prompt.
Muminpappa 🇪🇺🇺🇦@Life_is @jwildeboer "Ich will einen Computer, der macht was ich meine, nicht was ich sage!" ;-)
@[email protected]
Prof Wolfgang Wahlster war Gründungsdirektor des "Deutschen Forschungszentrum für künstliche Intelligenz" und hat dieses bis zur Rente geleitet. Er hat das Wort "Industrie 4.0" erfunden und galt jahrezehntelang als die deutsche Autorität für KI. Er hat den Begriff "Overanswering" ("Überbeantworten") geprägt, der kurz gesagt bedeutet: "Diese verflixte KI soll gefälligst tun, was ich meine, nicht das was ich sage!!1!!!"
@[email protected]
@Life_is @jwildeboer mir ist der Spruch erstmals Ende der Neunziger begegnet. Und ja, der passt auch auf KI 😂
Quinn Norton@jwildeboer "so, where is the documentation on effective and efficient prompting?"
@quinn "Prompting is more an art than a simple process we can document. For $LOTSAMONEY we can get you exclusive sessions with our resident prompt artists. Do note the NDA needed for that and you are only allowed to take notes, no recordings, no followup questions. This is a limited offer." ;)
lizbian@jwildeboer @quinn Soooo... Is someone already working on an LLM that will "democratise" prompting?
joui@jwildeboer We used to call computer instructions "commands" for a reason. Wishes are reserved privileges for sentient beings.
Ondřej Michálek@jwildeboer
And if it just so happened that you utterly embarrassed yourself by the atrocious quality of work your engineers produced, a little distraction might come in handy.
Quit while you're ahead 😉
Leszek@jwildeboer Have you tried reading their 245 page long "System Card" for Mythos? I went through the first 46 pages for now and it's painful how much fearmongering and how little actual on topic substance there is.
The 2. chapter is about risk assessment and there's everything but risk assessment, including a table which shows how much faster Mythos can be compared to their other models.
For me it looks like the usual pre-IPO marketing (Reuters and other news agencies suggest late 2026).
Alexandre Olivait's such a weird kind of religion... faith on an LLM's supposed artificial intelligence, prayers to the supreme beings in the form of prompts, misanswered prayers are the prayer's fault for praying wrong... this all checks out as a religion. but where's the profit?
asTra@jwildeboer
That's a Mythos 🤭
That's a Mythos 🤭
Alex o_O@spudtnik64 @jwildeboer All LLM products, since the first release of ChatGPT at least and perhaps earlier.
Remember that the first versions couldn't access the Internet because they weren't sure it was safe, so it had to be sandboxed - for the safety of humanity! Who knows what can happen if this BEAST is released to the public!
John Rohde Jensen@jwildeboer That and the old standby; make them sick - make them healthy
StarkRG@jwildeboer With one caveat: medication. Don't get me wrong, I'm no fan of the pharmaceutical industry, but s substance that has the potential to be medically advantageous also has the potential to be medically dangerous. Often, the difference between medicine and poison is a matter of dosage. Anyone who tells you a medicine is perfectly safe either don't know what they're talking about or are just flat-out lying.
JuanjoSan@jwildeboer What does it mean NDA?
@JuanjoSan Non-Disclosure Agreement.
Sheldon@jwildeboer wasn't only like 6 months ago that #OpenAI pulled a similar stunt with one of its project leaders "quitting in protest" in the name of Pandora because he was so terrified their technology was on the verge of achieving AGI sentience?
Bruce Simpson, Ph.D.@sysop408 @jwildeboer Oblique reference to known whackjob Leopold Aschenbrenner detected.
@bms48 ha, was that his name? When I heard that story, I just rolled my eyes and got back to my day... which actually involved talking a friend of mine down from believing the machines were about to make all knowledge workers obsolete within a year.
So I guess I didn't find that wacko to be funny.
@sysop408 @jwildeboer @Dianora @timnitGebru @timbray @pluralistic @emilymbender Do not sub out to a #GPT based #LLM that which #clangtidy (an established AST based tool for C/C++, much funded by #Apple prior to the #GenAI #lemmingrun ) is perfectly capable of doing without #tokenmaxxing https://clang.llvm.org/extra/clang-tidy/ Save the tokens and API logins for internal QA adversary activity
@sysop408 Yes. On the top 10 list of shady marketing tactics "drama sells" comes right after "sex sells", I'd say. On position 3 I'd put "artificial scarcity".
@jwildeboer you would think that if this tech is that blisteringly smart, the first thing it would do is figure out how to pay for its own existence. Like my cute little chatbot, you're not gonna conquer the world until you can afford to move out of your dad's basement into your own flat.
The gallant knight@jwildeboer exactly my thoughts. The FAZ even posted an article on it: https://www.faz.net/aktuell/wirtschaft/unternehmen/claude-mythos-von-anthropic-eine-ki-von-der-jeder-hacker-traeumt-200718701.html?premium=0xd91538c3de035243d728c04b902e954ab05ff0bad6d6c384525bd596b5381ba5
They totally fell for it it seems
They totally fell for it it seems
Claude Mythos von Anthropic: Eine KI, von der jeder Hacker träumt
Anthropic hält seine neue Künstliche Intelligenz für so gefährlich, dass es sie unter Verschluss hält. „Claude Mythos“ könnte „verheerende“ Auswirkungen auf die IT-Sicherheit haben. Was hinter den Sorgen steckt.