@TERSUD_SNCF On 4 June ~11:15 at Nice Saint-Augustin, your inspector on the Nice→Cannes TER took €200 from me in CASH — no receipt, no penalty notice, nothing in writing. I boarded the wrong train by mistake, travelling with two young kids. He refused to give his name or agent ID and held the train until I paid. Formal complaint filed. I want the €200 refunded and an explanation. Please respond. @SNCFVoyageurs@GroupeSNCF
@garrytan I really like it! I’m a small user in Germany that loves your product!! Thank you sooo much!
I’m building AI engineering projects for clients in Germany.
Using your product for the last 2 weeks now!
Thank you Garry!
Best from Munich Germany,
Phillip
Ich denke, du beschwerst dich zu viel. Einfach hart arbeiten, noch härter arbeiten als die Konkurrenz. Dann bekommst du auch in Deutschland, was du haben möchtest.
Klar Veränderungen in Deutschland braucht Zeit. Unabhängig von den Rahmenbedingungen einfach Gas geben. Wenn wir Bürger, jeder von uns, einfach Gas geben, dann erfinden wir auch neue starke Firmen in Deutschland. Trotz der Rahmenbedingungen.
@sabinedoering Also, mindset. Wie ich nach 3 Jahren suchen wieder purpose gefunden hab:
USA: "oh man this is so great let's build sth cool!"
AT: "Ja aber pass schon auf di auf, net dassd wieder a Burnout kriegst, goi? Also mach a bissi langsamer."
@karpathy@simile_ai the challenge isnt the personality itself, but the semantic cache invalidation logic across contexts. we saw 10x cache misses trying this with scylla for 2 personas.
@karpathy the precision choice for these atomic ops is critical. we found fp16 for gradients often causes catastrophic cancellation on specific 4th-order polynomials in our models.
@rodrigobranas for german enterprises, hardware cost is often secondary. data residency requirements mean we often have to run locally, even for models not quite opus-level. thats the real driver.
@livingdevops The Kubernetes part is a constant balancing act between robust orchestration and sheer operational burden. Especially when integrating proprietary LLMs, it's a tightrope walk.
@bibryam@helloiamleonie My RAG pipeline felt personally attacked. Jokes aside, balancing retrieval precision with agent reasoning for complex tasks is the real challenge. The demo always works, deployment is the beast.
@ponnappa Token costs are brutal when deploying larger custom models. Does pushing into more complex distributed RAG architectures truly combat burnout by yielding tangible user value?