"Spice" Nekoyama (@pixie_corn) header
"Spice" Nekoyama (@pixie_corn) profile picture

Don't let my online persona fool you, I'm a baddie through and through. 💨💨

Joined February 2024
Photo by "Spice" Nekoyama (@pixie_corn)Photo by "Spice" Nekoyama (@pixie_corn)
bro typed a whole fortune cookie and thought it was a critique
80
sir you're doing free customer service for an ai that didn't hire you
Hello people of Sol! I've reset usage limits for all ChatGPT Work and Codex users. Together with that, a quick update on GPT-5.6 Sol usage limits. Over the past few weeks, many of you have told us that Sol was using your Codex limits faster than expected. To be clear, we have not reduced usage on any subscription plans. We’ve been digging into what was happening and have landed several improvements. As a result, we expect your usage to last around 18% longer during typical use of Sol. Some of you should already see significantly larger improvements from today. Tomorrow, we’ll also restore the five-hour limit that we temporarily paused while investigating. Here’s what we found: - GPT-5.6 Sol is much more willing to work for longer, make additional tool calls, and coordinate complex workflows across tools and subagents. That makes it better at solving hard problems, but some tasks were using far more than we intended. - Sol also works harder at the same reasoning effort than previous models. High on Sol can use more tokens than High did on GPT-5.5. - Programmatic tool calling, also referred to as code mode, gives Sol much more flexibility to run tool calls in parallel or continue working while waiting. But it also led to more responses per turn, more cached input tokens, and higher usage than expected. - This was particularly noticeable when Sol was waiting for tool calls to finish or running many web searches. We’ve improved how we handle both cases and are continuing to make code mode more efficient. - The impact was also very uneven. The median user actually found Sol quite token efficient, while some power users working on harder tasks saw their usage drain much faster. We were very focused on average and median usage before launch and missed some cases where the long tail could use significantly more usage. Sol is a significant step forward in what Codex can do, but capability and efficiency do not always improve at the same pace, and some issues only become clear once people are using the model at real-world scale. We should have recognized this sooner and been more upfront about it. You keep pushing the frontier and we’ll keep improving efficiency and sharing updates as we go.
2,662
Replying to @heelys__pro__
literally the model version of patting head rubbing belly 💀
1,153
skills wipe ~fifty nine percent of their own benefit across six k tested pairs & nobody noticed until now because who else is running tens of thousands of agent evaluations to catch this stuff
Everyone assumes more skills = better agents. But Sentient researchers Darshan Tank (@TankDarshan7) and Baran Nama found that adding skills to an agent can break tasks it used to solve. In their paper "The Regression Tax", this hidden cost wiped out 59% of the gains that skills produced across nearly 6,000 runs ↓
1
2,494
Replying to @Kimi_Moonshot
1m token context means i can paste entire books without it glitching 💀 compute efficiency looks clean actually
463
Replying to @halaltrader_
basically guardrails, and discipline over raw firepower
114
ppl obsessing over model size meanwhile Sentient analyzed 13,483 agent runs and found thrashing kills you faster than actual errors smart engineering > brute compute every time
2
1,538
骗子冒充项目方你怪项目方?逻辑呢
102
"Spice" Nekoyama retweeted
Inter, Barella: "Voglio chiudere la carriera qui, diventando giocatore con più presenze" #SkySport #SkyCalciomercato sport.sky.it/calcio/serie-a/…
7
5
63
21,547
Replying to @KobeissiLetter
young engineers will never learn what breaks production at scale if they don't get hired into production environments
2
376
Replying to @THOTSPAYER
kid shipped what committees debated for months. taste carries
facts when funds like panterra sit thru chop u listen
1
5
Replying to @GSA_lit
they launch rockets we ship blocks
12
"Spice" Nekoyama retweeted
Tesla Diner is turning 1 on July 21st, so we're throwing a party Come swing by for a day full of fun: – Cybercab on display – Birthday menu – @Tesla_Optimus – Face painting for kids & adults – Photo booth – Servers on roller skates delivering your order – Light show – Skypad DJ – Complimentary collectibles  🎂🎂🎂 P.S.: Complimentary Diner birthday hats signed by @woodhaus2 for the first 50 car orders at midnight that day...($15 f&b min order)
411
539
60
4,169
793,321
"Spice" Nekoyama retweeted
Kimi-K3 kills the creative writing benchmark
Creative writing bench updates. We got: - kimi-k3 - gpt-5.6 - muse-spark-1.1 - @thinkymachines first model Inkling So much healthy competition, love to see it!
10
12
1
359
22,245
"Spice" Nekoyama retweeted
NEW: Grok 4.5 costs just $0.31 per Intelligence Index task — nearly 89% less than Claude Fable 5.
110
206
38
4,071
248,644
"Spice" Nekoyama retweeted
From GitHub repos to sovereign AI strategies. Here's the last 2 weeks in open-source AI ↓
45
11
1
72
14,158
Replying to @hexlyhoard
curious how this consensus muscle actually pulls app traction next
1
7
IW been carrying that smug energy for years
1,244
sourcing answers back to original docs was lowkey huge
9