FINRA is cutting the retake wait on its exams, SIE included.
From Jan 19, 2027:
β’ 1st or 2nd fail: 15 days (now 30)
β’ 3rd fail: 60 days (now 180)
Retake booked after Jan 19? FINRA says check if an earlier date helps (fees may apply).
Would 15 days change your plan?
@tundro agree. for me the big thing is how it handles long tasks: plan, edit, run tests, fix, without babysitting every step. what made it click for you, the tools or the model?
@Tony_Stef_ Good rule. Automated checks only test what somebody thought to test. A false line in the listing is exactly the thing no check catches, only a human reading it slowly.
@lenny_enderle Starting is almost free now, so the graveyard grows very fast. The real filter for me: does it have one real user yet. If not, it's still a demo, even when it's deployed.
@temitope_rotimi Same here π The second agent is good at catching the obvious stuff. The real problems hide in the code it didn't touch, so that part stays on me.
@Tony_Stef_ Classic one. The agent fixes the symptom very well, every time, because nobody asks it why the alarm exists. Honestly a noisy monitor is worse than no monitor, it teaches you to ignore alerts.
@valuecm Yes, this one bites hard. Idempotency key per intent, saved before the call, and on retry the agent gets the same result back. Without it the model just thinks 'no answer, try again' and you pay twice.
Sources: https://t.co/K0xrJR5GJx + /lcp, /inp, /cls and the optimize-lcp, optimize-inp, optimize-cls guides (Google)
Shipping a vibe-coded landing page? List it free on Truve with your build log: https://t.co/YyCGwBMA9b
Save this: Google grades your app with 3 numbers. Here are the passing scores.
LCP 2.5 s, INP 200 ms, CLS 0.1. And you must pass on 75% of real page loads, not on your laptop.
Honestly the easiest win is slide 4. A lazy-loaded hero image can quietly kill your LCP π
Which one is your app failing?
New weights just dropped: Hy4 vs Hy4 Flash
β’ Flash: 12B active of 220B. Hy4: 49B of 770B
β’ Both take 1M tokens of context
β’ Tencent's WorkIndex: Hy4 51.4, Flash 42.1, but Flash costs 1/20 per task
Which one are you shipping with?
@balon_f3 Very good point on VRAM, a lot of people miss it. 12B active makes it fast, but memory still pays for all 220B. Are you self-hosting MoE models or mostly on API?
@AxelGarciaK Same, long sessions are just more fun. The risky moment is when it compacts in the middle of a change and loses the "why". A short notes file with the current plan makes stretching much safer.
Source: https://t.co/8QKLJRsE9l (OWASP Top 10 for LLM Applications 2025)
Building an AI app? List it on Truve, free listing + build log: https://t.co/mLEn5DKysv
Save this: your AI app can be hacked with plain English. No code needed.
OWASP's Top 10 for LLM apps (2025), one risk per slide, with the fix from their own docs.
Honestly #9 surprised me: attackers publish real packages under the fake names AI assistants keep inventing π¬
Which one is your app weakest on?