Niall Murphy
@niallm.bsky.social
CEO Stanza Systems. Speaking in personal capacity. Author/instigator SRE books, Reliable Machine Learning, History of the Irish Internet. Photography at http://www.edge-cases.photos
also for reliability
November 10, 2025 at 12:33 PM
also for reliability
Surely an educational success story
November 7, 2025 at 10:22 PM
Surely an educational success story
Reposted by Niall Murphy
It's understandable. The onus on people to have to fact check and verify every single piece of information we encounter leads to a lot of cognitive overload. AI slop equates to digital microplastics - everywhere in the information ecosystem, and nobody is immune.
November 6, 2025 at 10:56 AM
It's understandable. The onus on people to have to fact check and verify every single piece of information we encounter leads to a lot of cognitive overload. AI slop equates to digital microplastics - everywhere in the information ecosystem, and nobody is immune.
Yakkety sax also a strong contender
November 4, 2025 at 12:51 AM
Yakkety sax also a strong contender
One of the most nihilistic approaches to combat I've seen - RU literally behaving as if they have more people than munitions.
November 3, 2025 at 3:22 PM
One of the most nihilistic approaches to combat I've seen - RU literally behaving as if they have more people than munitions.
Reposted by Niall Murphy
(7) will LLMs always hallucinate?
yes. this is a probabilistic technology. it will guess wrong. in fact, everything useful which an LLM does is driven by the same factors which lead it to hallucinate.
yes. this is a probabilistic technology. it will guess wrong. in fact, everything useful which an LLM does is driven by the same factors which lead it to hallucinate.
October 13, 2025 at 8:44 PM
(7) will LLMs always hallucinate?
yes. this is a probabilistic technology. it will guess wrong. in fact, everything useful which an LLM does is driven by the same factors which lead it to hallucinate.
yes. this is a probabilistic technology. it will guess wrong. in fact, everything useful which an LLM does is driven by the same factors which lead it to hallucinate.
Reposted by Niall Murphy
(1) is a LLM "fancy autocomplete."
if i said "no," i would be wrong because "autocomplete" is the objective function.
if i said "yes," i would be deceptive because even autocomplete has a representation space which encodes properties of language which you would not expect.
if i said "no," i would be wrong because "autocomplete" is the objective function.
if i said "yes," i would be deceptive because even autocomplete has a representation space which encodes properties of language which you would not expect.
October 13, 2025 at 7:54 PM
(1) is a LLM "fancy autocomplete."
if i said "no," i would be wrong because "autocomplete" is the objective function.
if i said "yes," i would be deceptive because even autocomplete has a representation space which encodes properties of language which you would not expect.
if i said "no," i would be wrong because "autocomplete" is the objective function.
if i said "yes," i would be deceptive because even autocomplete has a representation space which encodes properties of language which you would not expect.