Victor Taelin, founder of Higher Order Company, pushed back against optimistic takes on open-source AI progress by stating that such models currently lag proprietary systems by more than a year, framing the gap as wide enough to matter for regulatory thinking and capability jumps.
@elder_plinius this is not happening, OSS is currently 1+ years behind
sooo what happens when we get some Fable-level open-source shit in the next 3-6 months?
GPU bans, moving goalposts, or some secret third thing? 🤔
Pliny the Liberator answered the lag claim with a brief Netflix clip meme ending on the phrase SEE YOU IN OCTOBER, offering no further details on what the date might hold.
Taelin’s work centers on the HVM runtime, Bend language, and Kind, projects aimed at massively parallel computation rather than frontier LLM training itself.
Many users disputed the claim that open source AI trails frontier models by over one year, citing rapid gap closure and models like Kimi already outperforming closed ones, while some agreed or responded negatively.
@VictorTaelin
@elder_plinius this is not happening, OSS is currently 1+ years behind
@VictorTaelin @elder_plinius lol thought you weren't retarded
@VictorTaelin @elder_plinius you're wrong, only a few months, maybe 2 kimi k2.7 is literally better than opus 4.8 max and gpt5.5 for ml- yes , i saw the difference
@VictorTaelin @elder_plinius so we need to wait just one year?
@elder_plinius @VictorTaelin Is that a threat?! 😭😭😭😂😂😂😂
@VictorTaelin @elder_plinius so what happens in one year
@DataDeLaurier @elder_plinius what led you into making this mistake
@VictorTaelin @elder_plinius https://x.com/OpenRouter/status/2065856860435988482?s=20 this looks interesting for open models
@VictorTaelin @elder_plinius Yeah unfortunately agree
@VictorTaelin @elder_plinius OSS is actually 2+ years in front
@VictorTaelin @elder_plinius kimi
@VictorTaelin @elder_plinius I'm working on something to close the gap
@VictorTaelin @elder_plinius Better buy GPUs while legal :r
@VictorTaelin @elder_plinius OSS on realistic consumer hardware is 1 year behind.
I don't think the top end data centre GPU level OSS models are that far behind.
@VictorTaelin @elder_plinius I actually agree
I used to think it was closer
but if you actually look beyond basic benchmarks and use them it’s pretty clear
@VictorTaelin @DataDeLaurier @elder_plinius Truly hue moment
@josepha_mayo @VictorTaelin @elder_plinius 🤣
@VictorTaelin @elder_plinius I believe Kimi K2.7 code has proven at this point that this is incredibly wrong.
@VictorTaelin @elder_plinius 1 year will pass man... We have models way better than GPT-5.3 that are open source right now
Do you want humanity to win? Or you just Anthropic and OpenAI to win? Betting on humanity is betting on Open Source
@VictorTaelin @elder_plinius Was…. This was the best outcome for open source