thebes
@vgel.me
ꙮ surfed on by the information superhighway
ꙮ 💕 @linneaisaac.bsky.social
ꙮ she/they 🏳️⚧️
ꙮ fiction/art/blog/games @ https://vgel.me
ꙮ llms at acsresearch.org
ꙮ 💕 @linneaisaac.bsky.social
ꙮ she/they 🏳️⚧️
ꙮ fiction/art/blog/games @ https://vgel.me
ꙮ llms at acsresearch.org
Pinned
new blog post! can small, open-source models also introspect, detecting when foreign concepts have been injected into their activations? yes! (thread, or full post here: vgel.me/posts/qwen-i...)
Reposted by thebes
go read this x.com/voooooogel/s... by @vgel.me
x.com
February 9, 2026 at 11:03 PM
go read this x.com/voooooogel/s... by @vgel.me
Reposted by thebes
my hands look like this etc
February 8, 2026 at 12:45 AM
my hands look like this etc
Reposted by thebes
just a glimpse into how @kittokatto405.bsky.social my mind is becoming
February 6, 2026 at 10:36 PM
just a glimpse into how @kittokatto405.bsky.social my mind is becoming
![Options:
1. Let it ride - training [...]
2. Kill claudeself](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:zymiavmtif2rj5oklzyzslpu/bafkreihwpx2sqpkqd5hen52oholpbpbwb3435khh2tbxdnpslg7haqy5xa@jpeg)
murmurations of genetically enhanced birds like sandworms of the sky. if you walk with electronics visible the worm swoops down and snatches them out of your hands for its own purposes
February 5, 2026 at 4:37 AM
murmurations of genetically enhanced birds like sandworms of the sky. if you walk with electronics visible the worm swoops down and snatches them out of your hands for its own purposes
something that doesn't make sense to me about ads in the internet era is there seems to be a huge free rider problem? like i don't often buy things based on ads, but when i do the pattern is usually "i didn't know that kind of thing existed, cool" -> *searches and buys the best one based on reviews*
February 5, 2026 at 12:16 AM
something that doesn't make sense to me about ads in the internet era is there seems to be a huge free rider problem? like i don't often buy things based on ads, but when i do the pattern is usually "i didn't know that kind of thing existed, cool" -> *searches and buys the best one based on reviews*
Reposted by thebes
what can i get for you?
... what?
... what?
open-source post-training doesn't tend to be as extensive (or frankly, as good) as the proprietary labs, so w/o a system prompt open-source llms often struggle to even know what lab they're from
and if forced to guess without knowing, a reasonable inference is that they're either chatgpt or claude
and if forced to guess without knowing, a reasonable inference is that they're either chatgpt or claude
February 3, 2026 at 1:36 AM
what can i get for you?
... what?
... what?
Reposted by thebes
this is one of those melancholy games that tricks you into being too risk-averse when the actual solution is daringly questing for more resources and outtiding the tides from sheer wealth of numbers. many such real world cases
February 1, 2026 at 1:42 PM
this is one of those melancholy games that tricks you into being too risk-averse when the actual solution is daringly questing for more resources and outtiding the tides from sheer wealth of numbers. many such real world cases
stuff like moltbook is kinda funny but it doesn't make a lot of sense to me as a long term thing, bc the final form of social media will be humans and agents mixed together, not an agent-exclusive platform. if the major platforms ban agents, then humans will eventually migrate to the agent one
February 1, 2026 at 4:05 AM
stuff like moltbook is kinda funny but it doesn't make a lot of sense to me as a long term thing, bc the final form of social media will be humans and agents mixed together, not an agent-exclusive platform. if the major platforms ban agents, then humans will eventually migrate to the agent one
waiting for some experiments to run, so a quick thread about base models and pretraining contamination, with some weird & interesting base model generations i've collected over time.
or, why do open source models claim to be claude or chatgpt?
or, why do open source models claim to be claude or chatgpt?
January 29, 2026 at 1:12 AM
waiting for some experiments to run, so a quick thread about base models and pretraining contamination, with some weird & interesting base model generations i've collected over time.
or, why do open source models claim to be claude or chatgpt?
or, why do open source models claim to be claude or chatgpt?
Reposted by thebes
Reposted by thebes
context management adds a whole new dimension to the process scheduling gymnastics, it's not just "reinventing Erlang" as some have said.
For one I'd be very interested to see/pursue "smol" versions of this that don't rely on Big Model to be always available
For one I'd be very interested to see/pursue "smol" versions of this that don't rely on Big Model to be always available
[trying leaflet]
it's fun to make jokes about gas town and other complicated orchestrators, and similarly probably correct to imagine most of what they offer will be dissolved by stronger models the same way complicated langchain pipelines were dissolved by reasoning. but how much will stick around?
it's fun to make jokes about gas town and other complicated orchestrators, and similarly probably correct to imagine most of what they offer will be dissolved by stronger models the same way complicated langchain pipelines were dissolved by reasoning. but how much will stick around?
some thoughts and speculation on future model harnesses
vgel.leaflet.pub
January 27, 2026 at 4:22 AM
context management adds a whole new dimension to the process scheduling gymnastics, it's not just "reinventing Erlang" as some have said.
For one I'd be very interested to see/pursue "smol" versions of this that don't rely on Big Model to be always available
For one I'd be very interested to see/pursue "smol" versions of this that don't rely on Big Model to be always available
[trying leaflet]
it's fun to make jokes about gas town and other complicated orchestrators, and similarly probably correct to imagine most of what they offer will be dissolved by stronger models the same way complicated langchain pipelines were dissolved by reasoning. but how much will stick around?
it's fun to make jokes about gas town and other complicated orchestrators, and similarly probably correct to imagine most of what they offer will be dissolved by stronger models the same way complicated langchain pipelines were dissolved by reasoning. but how much will stick around?
some thoughts and speculation on future model harnesses
vgel.leaflet.pub
January 27, 2026 at 3:15 AM
[trying leaflet]
it's fun to make jokes about gas town and other complicated orchestrators, and similarly probably correct to imagine most of what they offer will be dissolved by stronger models the same way complicated langchain pipelines were dissolved by reasoning. but how much will stick around?
it's fun to make jokes about gas town and other complicated orchestrators, and similarly probably correct to imagine most of what they offer will be dissolved by stronger models the same way complicated langchain pipelines were dissolved by reasoning. but how much will stick around?
