Discussion
Every time there's a new study on the negative impact of LLM tools on education, work, communications, and media, at least half the commentariat will always go "but you shouldn't throw the baby out with the bathwater, 'AI' has constructive uses!" even as they pretend to be 'critics'.
How bad do the effects have to get before they finally go "okay, yeah, maybe it is more bad than good"?
@baldur Some people are unable to process negative externalities weighed against immediate positive benefits until it personally affects them. Or competing demands require prioritization.
If saying "thing is bad" doesn't work then we'd do well to find another strategy, such as understanding what the positive benefits they think they get and figuring out alternate solutions for those.
If it's about productivity or reliance on making money then making life/health less dependent on having a job.
@baldur to that I always respond "DDT also 'worked', asbestos also 'worked', single-use plastics also 'worked'"
LLMs are bad even if they "work", but eventhat is still disputed.
@baldur sure asbestos and CFCs have some negative impacts but they are really useful in some places
Install Fedizen
Get the full app experience
@baldur
... more bad than good? Who shall be the judge of that?
This Internet thing, wandering blithely around the world with no security in the Internet Protocol,
No authentication of source addresses. Anyone can put any source address in a packet. That made spoofing trivial, and it underlies reflection and amplification DDoS to this day.
Robert Morris’s 1985 paper on TCP sequence number prediction laid out the blind-spoofing attack, and Mitnick used it in 1994.
No confidentiality or integrity. Payloads and headers travel in the clear, and the IP header checksum only catches accidental corruption, not tampering.
Anyone on the path can read or rewrite packets.
Trust built into the layers above. The r-commands (rsh, rlogin) trusted by source IP address. Telnet and FTP sent passwords in plaintext. SMTP had no sender verification. All of it assumed the network was friendly.
Routing on the honor system. BGP, which glues the Internet together, accepts route announcements on trust. Prefix hijacks, like the 2008 Pakistan Telecom/YouTube incident, still happen; RPKI is a late and partial fix.
Fragmentation and ICMP abuses. Teardrop, Ping of Death and Smurf all came from the protocol working as specified.
Yeah, LLMs are just terrible. Gonna wreck human civilization. Well from where I sit, all this screeching about LLMs comes down to "DDT also worked".
And with this PHP shit-palooza underway, so did the Internet