Hi I like the simulation environments idea, it got me started working on a simulation based env for rail operations (experimental). I am trying to build railsimenv which works on dispatching trains on a real single-track Dutch line)
https://t.co/Er0OHksYPB
partly from it but also partly getting better at planning, like when a section is closed for an hour, do not queue trains in front of it, turn them back where they can be turned and keep the rest of the timetable intact.
But in some cases it made it a little worse turning back trains is not always okay, say sometimes the closure can be shorter like 20mins. The post trained model does not weigh in the length of the closure to make the decision it just learnt that turning back on each closure would be better.
Post-trained Inkling 2 small on Railsim. Inkling 2 Small started at 0.119 on the 45 hard evaluation tasks, below a simple rule-based dispatcher (0.180). RL taught it to stop running out of words: 0.186. Switching to the long context model increased the score to 0.259 before any training, and 0.316 after a second round of reinforcement learning.
Blog: https://t.co/IOaGNk1u6k
Space: https://t.co/y04dQN1BR5
@thinkymachines@tinkerapi
What kinda tasks look like this btw according to you? Like llms delegating gpus and training smaller models for some particular task/ benchmark performance ?
Was seeing around 1.5-3hr runs when I was using claude code(opus 5.5) to train small decision models using @sfcompute autoresearch for gpu delegation.
At @ScaleAILabs we created a high fidelity RL Environment of a trading desk in a real Texas battery grid.
GPT-6.1 Sol made $85,771 trading a real Texas battery. The battery's real operator made $20,647.
@adithya_s_k Nice work would love to know if one can contribute more envs to the repo. Trying to create a similar env for rail networks: https://t.co/GY88t2O8Sz
Hi I like the simulation environments idea, it got me started working on a simulation based env for rail operations (experimental). I am trying to build railsimenv which works on dispatching trains on a real single-track Dutch line)
https://t.co/Er0OHksYPB
We first shared these figures at last year’s opening ceremony.
It really feels like yesterday. Back then, we were meeting everyone for the first time. Weeks later, the names on our Zoom screens became project teammates and familiar faces in more conversations.
@bogoconic1, @nitin966, @furst_fly, @BpHigh_BpLow, @xhuydang, @mathlfs, @JunhoWon0, @Ming_Liang_Ang, @DmitryRybin1 and so many others from the first cohort, thank you for making those 12 weeks so memorable. Hope you still remember some of the long nights an days 😊
@khushbooverma_ Correction: FD interests are added in total taxable income so you kinda pay 30% tax + 4% cess.
If you want safe 7% investment rate with no taxes probably invest in PPF: https://t.co/OxKVmLs9Do
@Vishnu_Kaul Current gurugram tech hub only runs from some unicorns and a lot of relatively small firms. Kinda still a token Tech hub compared to other hubs.
@Vishnu_Kaul Also Bangalore has more offices of foreign Unicorn/ Mid tech(not as big as big tech but decently profitable foreign tech firms which are listed either on Nasdaq or respective home country exchanges and hence employ in decent numbers).
@Vishnu_Kaul And acquire newer land/ plan more Tech hubs in Gurugram adjacent cities like Panipat, Faridabad. The tech hub in Haryana is still nascent/ tokensake compared to Bangalore and even Hyderabad/Mumbai<>Pune.
few months ago there was a Turkish cloud AI conference where all speakers were men, many people said there's no expert women in the field and if there is, women should make their own conference 🧐 (I was quite baffled)
so bunch of Turkish engineers (@abilgegunduz@apatheticmagpie@aayseaktag) organized one with champions league speakers from companies like Cohere and @EDBPostgres sponsored it: @divaconference
wanted to give a shoutout, just great effort, wanted to be there but I wasn't available 🤗