@Teknium Sorry to hear that. Looks like Link had an amazing life. Hope you’re able to celebrate some of his best memories once the pain comes down
English
Rishabh Srivastava
4K posts
Rishabh Srivastava
@rishdotblog
Co-Founder @tryfactiq (YC W23)
Singapore Katılım Eylül 2011
1.4K Takip Edilen12.6K Takipçiler
Hey everyone. I haven't been very responsive on here the last week.
My dog, Link, who I've raised since he was a puppy over the last 13 years, passed away yesterday after being in the vet ER's ICU since last Wednesday for heart failure.
I put together some of my favorite pics of him to share so you all can see the most awesome animal friend I could ask for.
I'll be a bit slow probably through this week too, hope you all can understand 🙏
English
there's something magical about users finding your product in the wild and actually *using* it unprodded
hits so different, sp after years of top-down enterprise deployments
English
Rishabh Srivastava retweetledi
Launching factiq's open source plugin today!
It gives AI agents accurate macro data, market prices, and company financials in one place.
We have 25M+ series so far, including macro and trade data, US equity and fundamentals.
GIF
English
@rishdotblog Just tell fable to use codex for the grunt work. It will react with diabolical delight and start cracking the whip so hard on poor codex until the work is done. 😂
English
Anthropic's messaging: "Fable is amazing, I just keep run /loop for everything. Bye bye human SWEs"
Fable IRL, even when running in yolo mode: "I already delivered <small subset of work>. The rest is too hard. You do it."
Super annoying
English
@tulsani1 FWIW, I've never seen this issue with `codex --yolo`. Or even Opus in claude code. Seems like a Fable specific regression 🤷🏽♂️
English
@rishdotblog ChatGPT takes it to next level. It says it will do it but ghosts you at the end.
English
jic this shows up on the TL of anthropic engineers, submitted /feedback submitted, feedback id is 20492412-5c86-4bdc-96e9-c4a5e6337d65
English
@PradyuPrasad started off with diagnostics + an app for behavioral interventions!
English
@rishdotblog did they intend to sell medications?
English
been dealing with bad auto-immune issues for a few years now and remembered a startup that launched a few years to help with inflammation
they've now pivoted to B2B SaaS :(
wish consumer startup economics were less brutal
English
Rishabh Srivastava retweetledi
@rishdotblog text-to-sql ended up being the tarpit idea of the early LLM era
English
@rishdotblog Does it help if they have a catalogue that is well maintained?
English
outed 😅
it’s seriously hard. the tech is straightforward. trying to get people to tell you why they have 10 columns starting with revenue_version and when to use which is hard
Medha Basu@medha_basu
my cofounder literally has ptsd from trying to make enterprise databases work with LLMs
English
@shri_shobhit Yes for tech companies with reasonably clean schemas! No for legacy fortune 500s with columns like rxl_rte_7 😅
English
@rishdotblog True, data schema is the ultimate tribal knowledge.
Just curious, but having access to code, and an LLM analyzer would make it much easier to reverse engineer the meaning of each field? Backward compatiability and other linked data consumers might still be a blind spot though
English
@shri_shobhit YES! But a more fundamental (imo) was that enterprise llm stuff (specially for structure data) is best done in house. 3Ps have such little context - and often face a lot of internal resistance to get more context from data teams (who don’t want to get replaced)
English
@rishdotblog I feel the desire to cleanup database schema is the Marie Kondo PTSD of backend development :) it’s just an unending quest to get that perfect form of data
English
This is a great list. Hoping Sol isn’t nerfed for kernel/inference/datamix development in the way that Fable was!
Vaibhav (VB) Srivastav@reach_vb
I’ve got a ton of personal side projects I want to tackle with Sol: 1. Optimise MTP support in llama.cpp (to find low hanging fruits) 2. Have Sol figure out chat template mappings across various OSS runners and standardise to HF transformers 3. Bring back insanely fast whisper - it’s been abandoned for over a year! 4. Finish all the banks and financial connectors for my personal finance app (and ship itt) 5. Moar kernel optimisation!!!
English
@michalwols lol. way more "i'm right, the data must be wrong" behavior from claude code. codex (for all its limitations) understands epistemic humility
English
@rishdotblog lol it spent 20 minutes yesterday trying to prove hermes agent doesn't have 200k stars on github for me. it randomly stumbled on the repo and got completely distracted by the count, then started digging around to find out what was making github report that number
English
Claude thinks that recent Sandisk/Micron financials are "impossible". Every single time it starts working on a problem from scratch
Continual learning can't come fast enough
English
@rishdotblog To be fair, Gary Marcus and Michael Burry think it’s impossible also
English
@jietang Better creation and management of subagents via a harness (Claude Code or your own)
It's _okay_ at this right now, but can get better!
English
Rishabh Srivastava retweetledi
If you are on the verge of AGI or ASI, why isn’t your model smart enough to recognize espionage distillation in real time? You say “cure cancer in a few years.” Isn’t sniffing illicit distillation quite a bit easier than curing cancer? Why write letters to DC? Just use AGI.
English