@truth_archive

ARCHIVE of Interesting Perspectives and TRUTH From different sides, saved over the years! ⬇️

Joined March 2021
TRUTH ARCHIVE retweeted
An open-source AI model called Laya is beating TypeSafe AI's Jev in speed tests today. Here's what you need to know. Laya is a new open-weights "System One" decision model built by Convai Innovations, positioned as a free alternative to Jev. Like Jev, it doesn't generate text, it just picks from a set of typed options, but anyone can download and run its weights themselves. Developer account atomic.chat pitted Laya against cloud-hosted Jev in a Tetris match built with Grok 4.7. Laya won by making decisions 11 times faster, running entirely locally on a 16GB MacBook Air with no network calls. Other builders piled on fast. Mizorewww ported it to Apple's MLX framework as laya-mlx, running up to 50 times faster than Jev while using under 1GB of memory on-device. Developer Alex then ported it to CoreML, pushing 99.5% of the math onto Apple's Neural Engine for 3.7 millisecond decisions on an M5 Pro chip, released as FluidInference's FluidUse v0.2.0 on GitHub with weights on Hugging Face. Laya's own model card claims 83.8% accuracy versus Jev's 67.8% on its benchmark, and puts latency at 32.8 milliseconds versus Jev's 236 to 276 milliseconds, about 7.8 times faster. Some critics argue these game demos mainly reward low latency and that Jev may still lead on harder, non-algorithmic decisions. Key numbers: - 11x faster decisions in Tetris demo, on a 16GB MacBook Air - Up to 50x faster via laya-mlx port, under 1GB memory - 3.7ms per decision on CoreML/M5 Pro (99.5% ops on Neural Engine) - 83.8% vs 67.8% accuracy, Laya vs Jev, per Laya's model card - 32.8ms vs 236-276ms latency, Laya vs Jev Laya is fully open source and runs free on local hardware, while Jev stays a paid, cloud-hosted API.
Two AI agents just beat Minecraft's Ender Dragon in 8 minutes 43 seconds today. Here's what you need to know. Developer Ronak Malde, co-founder of Trajectory, paired a fast-reaction model called Jev with a reasoning model called Astra to play Minecraft together. Astra handled high level planning and wrote sets of possible actions, while Jev picked and executed moves with near instant, low latency decisionmaking. The run beat the Ender Dragon, the game's final boss, in 8 minutes 43 seconds, on peaceful mode using a fixed 1.16 speedrun seed. Malde open sourced the full code and harness setup on GitHub so others could replicate it. Within hours, another builder named Andy swapped in a different fast model called Jesse and beat the same boss in 7 minutes 6 seconds, on the first untrained try, for a cost of $0.00, using Malde's harness. Key numbers: - 8:43 run time, Jev + Astra - $0.97 total cost ($0.01 Jev, $0.96 Astra) - 7:06 run time, Jesse variant, $0.00 cost - 989.9K+ views on the original post Malde says the economics of Jev, tiny cost for fast decisionmaking, are what make this kind of real time agent movement possible.
1
2
6
841
TRUTH ARCHIVE retweeted
Local Laya moggs Jev at @grok 4.7-built Tetris 🧩 An open-weights System One model called Laya, beat cloud-based Jev at playing Tetris by making decisions 11 times faster, running locally on a 16GB MacBook Air! Run AI models locally -> atomic.chat
介绍比Jev快50倍,在你设备上跑的laya-mlx! 只在你的设备上占用最高1G内存 Laya是一个开源的类似于Jev的,基于文本输出概率的分类系统 我将其移植到MLX,并且做了一些性能优化! 视频中就是这个模型在我的本地M3Max上玩贪吃蛇 这个模型能够以每秒决策60次的速度玩贪吃蛇! github.com/mizorewww/laya-ml…
169
593
154
6,241
1,045,761
TRUTH ARCHIVE retweeted
. @modal moves mountains to make models manifest! Our man in pink @CompleteSkeptic will be speaking at Runtime, you won't want to miss it.
The full Runtime agenda is live, including talks from: @ScottWu46, Co-founder & CEO @cognition @CompleteSkeptic, Co-founder & CEO @typesafeai @dylan522p, Founder & CEO @SemiAnalysis_ @sarahookr, Co-founder & CEO @adaptionlabs @ajratner, Co-founder & CEO @SnorkelAI
12
10
2
183
53,250
TRUTH ARCHIVE retweeted
Stepping away while LLMs churn is the new “sorry, code’s compiling”. Jev makes instant feedback just a workflow away!
When I first saw JEV, as a former TikTok MLE, I knew modern recommendation and search systems could be revolutionized if it keeps evolving. JEV brings general-purpose semantic reasoning into a discriminative inference paradigm: arbitrary context in, calibrated structured decisions out. That matters because recommendation and search are exactly the kinds of decision systems that need to be both smart and fast. Times are changing. Here’s one example: with JEV + Perfectly, we use a complex query to understand AI researchers through their papers across a pool of 500 ECCV 2026 researchers, while achieving the same performance as one of Claude’s models.
38
21
3
594
94,689
TRUTH ARCHIVE retweeted
This can't just be transcript analysis, reveal your secrets! Brb playing with the live version @ElevenLabsDevs shipped
Realtime sentiment analysis with Jev and ElevenLabs. Each phrase takes on the color of the emotion it carries while the caller is still talking. Six meters on the right track the mood of the call.
35
43
5
778
115,926
TRUTH ARCHIVE retweeted
Replying to @other__reality
The researcher's life expectancy dropping every second 😭😭😭
2
6
216
11,257
TRUTH ARCHIVE retweeted
Replying to @other__reality
opus 5.5 ate with this one 👾
1
19
1,129
TRUTH ARCHIVE retweeted
I made this with one prompt using Opus 5.5 I spoke to my computer for 5mins, claude worked for 12 hours, and I woke up to this full prompt:
Claude Opus 5.5 has the best visual design of any model I have tested so far
484
880
505
11,048
3,928,648
TRUTH ARCHIVE retweeted
also it does look incredible nitter.cf/donaldjewkes/status/21…
I made this with one prompt using Opus 5.5 I spoke to my computer for 5mins, claude worked for 12 hours, and I woke up to this full prompt:
9
6
2
165
41,508
TRUTH ARCHIVE retweeted
I put a lot of heart into my technical writing, I hope it's useful to you all. 📌 Here's a pinned thread of everything I've written. (much of this will be posted on the Claude blog soon as well)
269
885
121
8,658
2,023,110
TRUTH ARCHIVE retweeted
we’re thinking of killing plan mode and using the shift+tab hotkey to adjust effort levels I don’t think the models need plan mode anymore, but if you’re a plan mode diehard would love to get your feedback on why
3,109
217
397
11,587
1,678,087
TRUTH ARCHIVE retweeted
Opus 5.5 designing LEGO 👀 I asked it to design a Microduck I can build with real LEGO pieces. It: > designed it life-size using 1113 real LEGO parts > verified: 3,204 connections, 0 collisions, every step buildable, centre of mass inside the feet 🤯 > made a 141-page LEGO-style booklet (237 steps) > priced every piece in the browser and prepared the orders on BrickLink
339
673
254
8,691
1,352,243
Replying to @typesafeai
This is not just about the signups - I logged into my account, payed for the service and eventually got rejected when trying to login to console.typesafe.ai My api key is also rejected so I guess you are happy to take the money but not to deliver the service?
3
2
27
5,488
🇮🇷|Western media omitted what Araghchi said to Kallas. The West reported only what Kallas "proposed" to Iran. But the truth lies in what Araghchi said to her. Araghchi told Kallas that the EU is "the greatest enabler of the crimes Israel commits against the oppressed Palestinian people in Gaza and the region." He added that the EU could have played a "more serious and effective role" in stopping the war but failed to do so. He explicitly pointed out: "The US repeatedly reneges on its promises and is the obstacle to diplomatic efforts." These were Araghchi's exact words. Western media did not report a single one of them. And what did Kallas say to Araghchi? She said: There must be "freedom of navigation" in the Strait of Hormuz. Who is blockading Hormuz? The US. The US targets 85% of Iran's oil tankers. The US has sunk five of them. The US has deployed Marines to the strait. Iran and Oman have established new shipping lanes, incorporating the southern channel into their territorial waters. Kallas does not demand that the US stop the blockade; she demands that Iran open it up. She said: End the US-Iran war. Who started it? The US. In February 2026, the US and Israel launched a 40-day war against Iran. Kallas does not condemn the aggressor; she demands that the victim cooperate in withdrawing. She said: Iran has seen an increase in executions and the use of the death penalty. Who killed 168 girls in Minab? The US. Kallas says not a word about it. This is the truth: Kallas did not come to negotiate. She came to deliver a message Washington's message. She is the EU's foreign policy chief, yet the EU's foreign policy regarding war and peace has never been decided in Brussels. It is decided in Washington. Araghchi saw right through her. That is why he "called on the EU to play a constructive role," this was not mere diplomatic rhetoric; it was an expression of contempt for a puppet. If the EU truly wanted to be "constructive," it should have: condemned the US for starting the war; demanded the US lift the blockade on Hormuz; recognized the new shipping lanes established by Iran and Oman; condemned Israel's genocide in Gaza; and stopped supplying weapons to Israel. But Kallas did none of these things. She did three things: demanded concessions from Iran, demanded concessions from Iran, and demanded concessions from Iran. She is not a diplomat; she is a mouthpiece for the European bureau of the U.S. State Department. And Araghchi was speaking to that mouthpiece.
171
2,094
157
4,514
120,767
🇮🇷|"We responded, but we did not hit civilian populations. They came into surrounding territories and targeted our universities, our healthcare centers, our schools, our hospitals." | Iran’s President Pezeshkian at the UN General Assembly He didn’t name the country. He didn’t need to. The country that bombed a girls’ school in Minab and killed 168 children. The country that hit a sports stadium in Lamerd. The country that struck universities, hospitals, clinics, and schools across Iran. And then called it “precision strikes.” Iran hit military targets. US and Israel hit civilians. That’s the difference. That’s the whole story. #Iran #UNGA #US #Israel #WarCrimes
3
10
1,140
TRUTH ARCHIVE retweeted
❗️ "They bombed a school": Iranian president accuses the US and Israel of killing innocent children with weapons supplied by them
🚨 Iran is not pursuing nuclear weapons & rejects limits on its peaceful nuclear program: President Pezeshkian at UNGA
3
7
379
As Iranian President Masoud Pezeshkian addressed the UN General Assembly, a US delegation walked out of the chamber. Pezeshkian accused the US and Israel of attacking Iran and rejected accusations that Iran is a terrorist state, stating, “We have been the victims of terrorism”. انسحاب الوفد الأمريكي من قاعة الجمعية العامة للأمم المتحدة بالتزامن مع اعتلاء الرئيس الإيراني مسعود بزشكيان المنصة، والذي وجه اتهامات لواشنطن وإسرائيل بمهاجمة بلاده، رافضاً اتهام إيران بالإرهاب بالقول: “لقد كنا ضحايا للإرهاب”.
44
89
5,507
TRUTH ARCHIVE retweeted
The President of Iran at the UN : “I come from an Iran where they bombed a school of children and they perished under bombs and weapons used by the US and Israel”
9
433
7
1,906
21,723
TRUTH ARCHIVE retweeted
Marco Rubio is spewing the same tired, desperate bullshit yet again. If Iran had developed nuclear weapons years ago, the US and Israel would have thought twice before treating the region like their personal playground. Nuclear deterrence is brutally simple: when you have the capability to hit back, these two vultures think twice before coming for you. This entire catastrophe had been avoided if Iran had possessed a credible nuclear deterrent from the start.
9
75
6
190
7,176