Angel Sirlopu Cumpa

Arquitecto de software / Developer
Papá de Thiago Nicolás

[Angel Sirlopu Cumpa @AnguelSC]

En el 2007 pedía prestada una PC estudiar a un familiar, en el 2010 mi madre hizo el esfuerzo de comprarme una PC que me duró muchos años, ahora tengo la oportunidad de gozar de un hermoso equipo y TODAS estas etapas tiene un factor en común. El apoyo incondicional de mi madre

[Remek Kinas @KinasRemek]

basal-omni 🇵🇱 - TOP1 🤩🤩🤩 🌍 na świecie w modelach otwartych oraz TOP2 wśród zamkniętych i otwartych

Wg niezależnego benchmarku modeli system-one gdzie testowane są wszystkjie modele na świecie - JevBench (link do bencha w pierwszym wpisie). Czyli można? Można! Polskie AI 🇵🇱

basal wyprzedził model OpenAI, Microsoft oraz pierwowzór z TypeSafe (jev).

Thank you, @airesearch12 , for the independent evaluation on JevBench.

[Nandakishor m @Nandakishorm1]

Vega Universal decision model. Now you don't need to train a separate model for each decision; you can use one engine and multiple LLMs as needed. This solves an important problem current decision models face: accuracy and calibration error. The idea is simple: you train a small projection layer and use that to attach any llm, vegaml package automatically ports that to a common dimension. That's it: joint training with invariance loss
https://t.co/HsjVaoTqTe
https://t.co/ai8SnOElQ8
https://t.co/UEdZMCbyuz

[Maziyar PANAHI @MaziyarPanahi]

BOOM!

130K RAG claims verified by Jev!

Happy Sunday.

https://t.co/N5YrXBaWhR

[Ren @Ryrenz]

🕹️ 谷歌刚开源了 ARTEMIS，能让 AI 像人一样拿着一台真 Android 手机点来点去。

8 月中旬才建的仓库，GitHub 上已经 1.2 万 star，一千三百多个 fork，在 Google Research 的 AndroidWorld 基准上跑出 99% 以上的任务完成率，那套题覆盖 20 多个应用、100 多个多步骤任务。

写过 Android 自动化测试的都知道那种折磨：一个登录流程要写几十行定位代码，界面一改控件就全废，脚本比功能本身还难维护。ARTEMIS 的路子是你用自然语言说要测什么，它靠元素索引去点，认不出来就退回坐标和视觉定位，跨应用的流程也能一路走完。

它原生带一个 MCP 服务，Claude Code、Windsurf、Antigravity 这类工具可以直接驱动手机，顺手把 Logcat 和截图收回来。赶时间就用每步三到五秒的快速模式，想跑长时间稳定性测试就换成执行前先核对目标、卡住了再交回来的那一档。另外它第一次在设备上跑任务时会装一个读屏幕布局的辅助服务，只在手机本地运行、不往外发数据，失败了就退回 UIAutomator2。

手机测试这件事，终于不用全靠人肉手指了。

GitHub：https://t.co/IAfR2RDHus

[Nandakishor m @Nandakishorm1]

How to use your own data to build a decision model without retraining and on any LLM with calibration
Colab Notebook To Try: https://t.co/bcJWF3hful

[Akshay 🚀 @akshay_pachaar]

LLMs vs. Jev, clearly explained!

TL;DR

The key difference is not that Jev generates faster.

Jev does not generate text at all.

A traditional LLM receives context and produces an answer one token at a time. Even when the output is a small JSON object, every token depends on those generated before it.

Jev receives the same context but evaluates predefined decisions directly. When those decisions are independent, it can evaluate all of them in parallel.

Consider an agent handling a failed deployment. It may need to determine:

→ Whether the incident is urgent
→ Which team should handle it
→ Whether the proposed command is risky
→ Whether the task is complete

An LLM generates a response containing these answers sequentially. The application then parses and validates it.

With Jev, you define the questions and expected answer types upfront. It evaluates them together and returns typed answers with probabilities.

Jev supports three decision primitives:

1. **Choice** selects from known options, such as engineering, billing, or sales.

2. **Score** places the input on an ordered scale, such as low, medium, or high risk.

3. **Noul** evaluates a yes-or-no condition and returns the probability that it is true.

The probabilities matter as much as the selected answers.

If engineering receives 91% probability and billing receives 9%, automatic routing may be reasonable. If the probabilities are 52% and 48%, the system can escalate, gather more context, or call a stronger model.

This keeps control inside ordinary software.

Code owns the thresholds and consequences. Jev supplies the semantic judgment that a normal `if` statement cannot derive from unstructured text.

It works best when the possible answers are known, the decision depends on meaning, and a careful person could judge the input quickly.

It is not designed for writing, summarization, code generation, arithmetic, or decisions requiring several dependent reasoning steps. Independent questions can run in parallel, but decisions that depend on earlier results must remain sequential.

Jev also cannot return an option outside the declared schema, but it can still select the wrong valid option. Type safety prevents malformed outputs, not incorrect judgments.

The clean mental model is this:

LLMs generate new language when the answer space is open.

Jev evaluates known paths when the answer space is bounded.

I wrote the full breakdown explaining Jev and where it fits.

The article is quoted below.

[Vox @Voxyz_ai]

Give Opus 5.5 𝗮 𝗰𝗹𝗶𝗰𝗸𝗮𝗯𝗹𝗲 𝗺𝗼𝘁𝗶𝗼𝗻 𝗱𝗶𝗰𝘁𝗶𝗼𝗻𝗮𝗿𝘆: 𝟭𝟮𝟬 𝗺𝗼𝘃𝗶𝗻𝗴 𝘀𝗮𝗺𝗽𝗹𝗲𝘀, each labelled with its name and a ready-made prompt. Click the ones you like, and Claude Code builds them into your video or site.

Five packs: motion graphics, camera moves and transitions, UI components, landing page and web effects, and photo looks and lighting.

Four steps:
1. Paste the line at the end into Claude Code to install the skill
2. Tell it "Open the motion-prompt library"
3. Tick the samples you like in the gallery and press "Send to AI"
4. Back in Claude Code, say "Edit the opening of this video like this"

It gets each sample's motion spec and reference code, and it works with Remotion, CSS or Canvas. Without the skill, press "Copy picks" and paste them in.

Paste this into Claude Code and it installs itself 👇
"Read https://t.co/QLB00Jm3Uf and install the motion-prompt skill in this environment. When it is installed, tell me how to use it in three lines."

[thameem @th2049]

convert files with just drag + shift

[KC @karanC_12]

Opus 5.5 vs DeepSeek V4.1 Flash-1017

•Opus: 20 mins, $6
•DeepSeek: 1 hour, $0.23

Opus cooked, but DeepSeek got surprisingly close at 26x cheaper.

[Guillermo Casaus @_guillecasaus]

👋🏼 ADIÓS A CAPCUT.

Ahora puedes convertir un vídeo largo en varios Shorts listos para publicar.

Y sí, lo hace automáticamente.

Se llama OpenShorts y es open source.

¿Lo mejor?

→ Encuentra los mejores momentos del vídeo
→ Genera clips en formato vertical
→ Añade subtítulos automáticamente
→ Prepara contenido para Shorts, TikTok y Reels

Todo dentro de una misma herramienta.

Ideal para reutilizar podcasts, entrevistas o vídeos largos sin tener que revisar y editar cada fragmento desde cero.

Repo en comentarios 👇

[Ozgur Guler @ozguramdin]

Ahenk'i yayınladık: açık ağırlıklı, Türkçe metinden sese (TTS) modeli.

Sayıyı, tarihi, fiyatı ve telefon numarasını doğru okuyor. "mı" ile biten sorularda sesi yükseliyor. Talimatla kızgın, neşeli ya da şaşkın konuşuyor, gülebiliyor.

Videodaki bütün sesler Ahenk'in. Sesi açın.

[Leon Lin @LexnLin]

earshot v0.2 is out now!

made a lot of changes and it can now turn a sound into code.

https://t.co/IQ9TLIvbiu

[Marco Franzon @mfranz_on]

Left: YOLO26n fine-tuned 50 epochs on PPE data

Right: stock YOLO26n finds people, a local 27B vision model (Jev) judges each crop. Zero-shot, no training.

This is probably the future of computer vision.

[Higgsfield AI 🧩 @higgsfield_ai]

DeepSeek V4.1 Flash looks like the new SOTA for motion design.

Look at the 3D detail and UI transitions.

Made with Higgsfiled.

[Tom Huang @tuturetom]

我去，DeepSeek V4.1 Flash 1017 做教程类动效视频已经这么牛了吗，感觉和Opus 5.5没啥区别了😂

我们用它在 OpenDesign 的 Motion Design 里做了一条教程，教大家怎么在其他 coding agent 里接入 OpenDesign API，用上它和其他模型，效果简直不要太好了

[taras @voidisomorphism]

built an arcade to play with the different system one providers out there

attention is all you need? here's how clef, jev and luna decide what they should pay attention too

clef find me boring... luna not

try it out at https://t.co/HSlCxjvTTg more in 🧵

[Greg Pstrucha @grichadev]

so yes, it stand very well against different vision models for people detection.

it's a bit of imperfect science, some of the tools for people detection are meant to be done on video stream, not on video frames, but the fact jev can do it on a single video frame, and with that level of accuracy for both finding people, matching their faces and recognizing known folks (i.e family members) is quite impressive.

[Xunzhuo @XunzhuoLiu]

Turn any camera into a stream of decisions.

Every second, d3-edge (0.6B) watched this rescue and answered 8 plain-English questions in ~0.1 s.

New questions, new scene: pet cam, bird feeder, doorstep. What would you ask?

https://t.co/GBVE6aUdz8

#Jev #Decision

[Maxime Labonne @maximelabonne]

Small decision models can be insanely good when fine-tuned.

We played with d1-omni-600M and it can destroy any model on the games + format (state, input) you train it on.

[Mehmet Eren Dikmen @ErenAILab]

Jev’in yeni güncellemesiyle birlikte artık realtime görüntü işlemek mümkün olabiliyor.

Önceden sadece bu videoda birisi var mı yok mu diye sorabiliyorduk.