@zb_bizness: @BeerLoga_RIGA #riga #zbbizness #latvija

ZB bizness 🇱🇻
ZB bizness 🇱🇻
Open In TikTok:
Region: LV
Saturday 29 August 2026 13:57:59 GMT
17443
367
16
188

Music

Download

Comments

vorona711
vorona :
Огре столица Риги.
2026-08-29 18:01:03
5
raimondsbumbiers
R.B.🇦🇹🇧🇻 :
bil i nepanravilas....soryyy
2026-08-29 18:13:09
0
lubasan5
Lubasan :
Мой Кенгарагс ,любимый🥰
2026-08-29 17:29:14
4
user91469600092615
Ромка :
так так заеду👍 выглядит очень аппетитно 👍
2026-08-29 19:09:25
1
vikaavdevich
vika.musipusi :
Вкусно 👍🥰
2026-08-29 17:28:08
1
olgadimper
Ольга :
где находится?🤩
2026-08-29 15:08:53
1
rat.chett
RatcheT :
повар нужен с опытом? живу в кенге
2026-08-29 15:05:19
2
karmenakoz
Karmenakoz :
а где именно кенгаракси.?
2026-08-29 19:52:59
0
oksanabeljakova
Oksana Beljakova :
супер
2026-08-29 19:31:19
0
paradoxlv
PARADOX :
darba laiks no pl 18.00???
2026-08-29 19:10:44
0
andrey.kochetkov1967
Andrey.Kochetkov :
ЗАЕДУ НА ДЕГУСТАЦИЮ ОДНОЗНАЧНО
2026-08-29 14:06:11
0
_alla650
Алла :
мне не понравился , был немного сыроват
2026-08-29 18:05:06
0
olama777
Olama777 :
Моьодцы ребята 👍👍 ,процветания вам 🥰
2026-08-29 16:48:10
0
astroanna
astroanna :
Живу рядом, и даже заходить не хочется, контингент так себе
2026-08-29 19:34:57
0
karmenakoz
Karmenakoz :
👍👍
2026-08-29 19:54:16
0
To see more videos from user @zb_bizness, please go to the Tikwm homepage.

Other Videos

Agent memory is not a bigger context window. It's a read-and-write system — and most people only build the read half. Wrong question:
Agent memory is not a bigger context window. It's a read-and-write system — and most people only build the read half. Wrong question: "how do I give the agent more history?" Real question: "what gets retrieved, and what earns the right to be stored?" One request: "Plan my Austin → NYC trip like last time." 🗂 FOUR STORES, not one blob. Working — the current task, this turn only. Episodic — what happened before. "Last trip: JFK" Semantic — durable facts. "Prefers aisle" Procedural — rules and skills. "Travel policy" 🔎 READ — the reader searches only the three long-term stores. Working memory isn't searched; it already holds the turn. Three cards come back. Everything else stays dark. 🧾 CONTEXT — the request on top, three retrieved rows underneath. That's it. Left out: hotel dates, home city, the 14-day rule, 900 other memories. 🛠 ACT — search_flights() → Morning · aisle · JFK. The action is downstream of what was retrieved, not of what the model "knows." ✍️ WRITE — three candidates hit a gate. "Flights < 10 AM" → SAVED. Durable. "Price today: $312" → DROPPED. True for an hour. "Loading status..." → DROPPED. True for a second. The survivor writes back into Semantic. 👀 The half nobody builds: the write gate. Retrieval gets the attention, but an agent with no gate saves everything, and an agent that saves everything retrieves garbage next week. Your recall problem in month three is a write-discipline problem from month one. The rule: memory is what survived a filter twice — once on write, once on read. Most "memory" is just the whole transcript re-stuffed into every prompt. That's not memory, that's a bigger bill. Send this to whoever is about to raise their context limit instead. 📸 Screenshot the last frame — four stores, one gate, the full loop. Follow @hackproduct — scary AI concepts, made shippable. ⚡ . . #agentmemory #AIagents #agenticAI #contextengineering #AIengineering

About