Harshit Chourasiya

Scaling RL / CEO at dedo-ai, open source bounty hunter , winner of many hackathon's, making humanity and AI better for tomorrow!

[Harshit Chourasiya @harshtech123]

@omouamoua agree , thats why for real interactions labs showing interest in the enterprises data that simulate real world !

[Harshit Chourasiya @harshtech123]

@JongwonPar9958 I know your going to building something great soon with or without YC doesnt matter :)

[Harshit Chourasiya @harshtech123]

@exploringfornow I think we should connect , we have more data that is valuable !

[Harshit Chourasiya @harshtech123]

Great opportunity for users to try @Replit

[Amjad Masad @amasad]

@OpenAI You can code with OpenAI models on @Replit for free.

Our router also makes high-end models incredibly cost-efficient.

If you’re a business looking for an independent, multi-model alternative to Cursor, we’d be happy to fund your transition.

[Harshit Chourasiya @harshtech123]

@divy93t Wants to see gemini to outperform in critical workflows.

[Harshit Chourasiya @harshtech123]

@jcyhc_ai @janhkirchner @jiaxinwen22 Prompt injection is one of major concerns right now and this much improvement with AAR tells us that what ai can actually improve in security.

[Harshit Chourasiya @harshtech123]

Now with updated version , new tasks will come up ! excited to see how agents will perform with each matrix

[Ryan Marten @ryan_marten]

We've pushed a version update to the Terminal-Bench dataset and leaderboard.

Terminal-Bench 4.0 calibrates task resources (time, cpu, memory), implements task fixes, and removes saturated tasks.

[Harshit Chourasiya @harshtech123]

@katelyn_lesse i am very excited to see what the team will comes with next in hardware standard !

[Harshit Chourasiya @harshtech123]

@StevenDillmann welcome steven , can we have chat for sometime !

[Harshit Chourasiya @harshtech123]

@wenhaocha1 yes there is lot , for that we will still needs somekind of benchmarks .

[Harshit Chourasiya @harshtech123]

very good benchmark , we are also working on benchmarks for the AI ecosystem , we are doing research part for that and going to come up with something useful next month !

[Steven Dillmann @StevenDillmann]

We're releasing Terminal-Bench-Science: a benchmark for evaluating AI agents on research workflows across scientific domains.

An ongoing Stanford-led community effort, built by the team behind Terminal-Bench together with scientific domain experts at research institutions worldwide. v0.1 has 70 tasks. Claude Opus 5 solves only ~30%.

1/n 👇

[Harshit Chourasiya @harshtech123]

@Nishant_N_25 It will soon :)

[Harshit Chourasiya @harshtech123]

@maxccummings I think we should connect for that gap , we have enterprise data and still figuring out what to do with that !

[Harshit Chourasiya @harshtech123]

@ryan_marten We are doing this and spending a lot of time and efforts to make evals quality better .

[Harshit Chourasiya @harshtech123]

@shengkunye @Tiny_Fish Wow interesting ! But how can we have such agent for free .

[Harshit Chourasiya @harshtech123]

@ScaleAILabs Great benchmark i support this too .

[Harshit Chourasiya @harshtech123]

@neversupervised These two patterns makes sense :)

[Harshit Chourasiya @harshtech123]

@andrewho03 Even if never worked big labs but has good connection over there , it also applies!

[Harshit Chourasiya @harshtech123]

@mjlbach Seriously too real :)