RasG

4.1K posts

RasG banner
RasG

RasG

@Reportooor

They say knowledge is power, but sometimes a little misinformation goes a long way. 📝Editor @DisruptAfrica

Katılım Aralık 2023
273 Takip Edilen107 Takipçiler
RasG
RasG@Reportooor·
@SentientAGI That 7 error classification applies to legal in addition to medical reasoning too. Outdated info in case law, conflicting sources in studies.
English
0
0
1
436
Sentient
Sentient@SentientAGI·
New paper is out—CryptoAnalystBench: Failures in Multi-Tool Long-Form LLM Analysis. Improving reasoning agents starts with understanding how they fail. We study the errors that emerge when agents answer open-ended questions with no clear answer. Check core failures & fixes below 🧵
Sentient tweet media
English
41
13
128
14K
RasG retweetledi
Cointelegraph
Cointelegraph@Cointelegraph·
🚨 JUST IN: Bhutan reportedly selling $30M in Bitcoin per Arkham data.
Cointelegraph tweet mediaCointelegraph tweet media
English
80
49
241
26.2K
RasG retweetledi
Cointelegraph
Cointelegraph@Cointelegraph·
🚨 HUGE: $14.16B in $BTC options set to expire Friday, nearly 40% of Deribit OI with max pain at $75K.
Cointelegraph tweet media
English
117
107
456
58.5K
RasG retweetledi
Crypto Rover
Crypto Rover@cryptorover·
HUGE Liquidity below Bitcoin! We all know what this means...
Crypto Rover tweet media
English
134
64
516
90.3K
RasG
RasG@Reportooor·
@kevinroose so we're tracking who burns the most tokens for closed companies as a status game? feels like bragging about gas usage in someone else's car. curious if tokenmaxxers have checked what @SentientAGI is doing with decentralized compute
English
0
0
0
68
Kevin Roose
Kevin Roose@kevinroose·
One OpenAI employee used 210 billion tokens *last week*. A single Claude Code user spent $150,000 in a month. Meta, Shopify and other companies now factor token use into performance reviews. Are you tokenmaxxing, anon?
Kevin Roose tweet media
English
17
9
99
14.8K
RasG
RasG@Reportooor·
@AssiniKojo @Coredao_Org "no tech and product" meanwhile the code is open source on github and the chain has been running since 2023 with zero downtime
English
0
0
0
15
🆓
🆓@AssiniKojo·
@Coredao_Org You guys are fools. DO YOU EVEN UNDERSTAND THE ENGLISH TERMS YOU GUYS USE. HOW MORE UNDERSTANDING OF BLOCKCHAIN. You guys are thieves you don't have any tech and product. Nothing have ever worked for you except stealing with big terms you don't understand
English
1
0
4
119
Core DAO 🔶
Core DAO 🔶@Coredao_Org·
Core is becoming the quantum-resistant Bitcoin layer. Cryptography doesn't stand still. Neither does Core. Here's what's being built. 🧵
Core DAO 🔶 tweet media
English
178
309
857
80.8K
RasG
RasG@Reportooor·
@br_ttany @Sasha_Nuova_AI yo this is sick... if you want sasha to actually own her earning and scale without babysitting check out @SentientAGI they're building open source infra for autonomous agents to monetize and operate decentralized
English
0
0
0
9
Joan
Joan@02znin100887·
@richrines Sub second finality?
English
1
0
2
267
richrines.core 🔶
richrines.core 🔶@richrines·
Tomorrow we're announcing a major technical upgrade coming to Core. Very excited for you all to see. 🔶
English
105
142
522
47K
RasG
RasG@Reportooor·
@Coredao_Org Designing for holders, traders AND institutions means Core isn't splitting its market, its capturing the entire BTC holder spectrum
English
0
0
2
862
Core DAO 🔶
Core DAO 🔶@Coredao_Org·
Conservative holders need safety. Yield seekers need performance. Institutions need compliance. Core built products for each. Different user, different product, same requirement: CORE. Segmented by design. Unified by demand. 🔶
Core DAO 🔶 tweet media
English
127
237
701
33.8K
RasG
RasG@Reportooor·
@tripathi_neil @eonsys The fact that no public bridge existed between the brain and body until now is actually crazy, this is real infra work
English
0
0
0
20
Neil Tripathi
Neil Tripathi@tripathi_neil·
This project has been fascinating. In my last tweet, I said I wanted to test if a brain could do better under structural vs plastic updates. To do this, I had to: 1. Recreate what @eonsys did 2. Figure out which circuits do what, and are actually worth updating For (1), I reverse engineered a closed-loop bridge between the FlyWire connectome and a physics-simulated fly body. The brain and body are open source, but no publicly released bridge exists between them, so I created it and am hoping to open source that soon. For (2), I dug through a lot of prior literature on what different neurons and pathways are thought to do, then ran targeted ablations directly on the connectome: silence specific neural pathways and see what changes in behavior. In one case, silencing just 210 neurons out of ~138,000 significantly disrupted escape responses to looming from one side. That led to a broader map of how sensory signals reach descending motor output. One thing that surprised me was how closely the model matched evidence from prior literature. What surprised me even more was discovering that by the time signals reach the descending layer, vision, smell, temperature and humidity are far more segregated than I expected. Some modalities stay on remarkably distinct routes. One especially interesting case was DNb05. It is not a dedicated thermo/hygro (temperature/humidity) neuron. It is known as a multimodal steering-related neuron with strong visual input. But in this bridge, it is the dominant direct descending route by which temperature and humidity signals reach the motor readout. Remove this, and that seemingly unrelated neural pathway collapses, while the other major pathways stay intact. Marin et al. (2020) noted that hygrosensory signals may reach motor circuits in as few as two synapses, and my simulation identifies the specific descending neuron and confirms it causally. Not everything gets mixed together into one pooled representation like in many modern neural networks. Before you update the brain, you need to know which circuits are flexible and which ones are load-bearing. Seriously cool to see the wiring be this clean.
Neil Tripathi tweet mediaNeil Tripathi tweet media
English
6
10
68
7.7K
RasG
RasG@Reportooor·
@SentientAGI timeline mentions up, discord buzzing, builder apps flooding in at 500+, and now Arena goes live. the momentum is really fucking real.
English
0
0
1
998
RasG retweetledi
orbitals
orbitals@orbitals_gg·
New limited-time reward pool is live in the Closed Beta 🪐 Head to the DeFi platform and make your call: Will ETH close above $2100 on March 15th (ET)? Pick your side and lock it in! Correct calls will share the 100 USDC reward pool. Join the platform to make your prediction in the link below
orbitals tweet media
English
14
26
186
36.1K
Core DAO 🔶
Core DAO 🔶@Coredao_Org·
What would change your daily Bitcoin experience the most?
English
113
140
425
26.1K
RasG
RasG@Reportooor·
@tripathi_neil This makes sense, thank you for making it clearer
English
0
0
1
15
Neil Tripathi
Neil Tripathi@tripathi_neil·
Great question on open vs closed on VB. Opens range 0.419-0.505 overall vs closed 0.476-0.728, so there is actual overlap at the bottom. Gemma 3 12B (0.505) beats Claude 3.7 Sonnet (0.476) outright. InternVL3-8B (0.445) is close too. Big gaps on MEFR (minimal-edit faithful flips) and ToMAcc (perspective reasoning). Opens much weaker there, and SelRank (confidence calibration) is near zero - basically random. Qwen3-VL-8B (0.419) weakest by far, which was surprising to me initially, but was explained by 50 abstentions + 66 unparsable outputs total. Format adherence was a disaster. Opens are catching up fast on aggregate scores (even topping some mid-tier closed), but frontier closed still win on precise evidence-tracking & calibration. Gemma 3 12B looking strong though!
English
1
0
3
1.6K
Neil Tripathi
Neil Tripathi@tripathi_neil·
Just submitted my first paper to arXiv, and I found something that fits the growing conversation around newer models hedging their bets more and more. VB: Visibility Benchmark - checks if vision-language models can apply common-sense reasoning to determine what's actually visible in a photo. Joint work with Ernest Davis at NYU. 9 models tested: GPT-5, GPT-4o, Gemini 3.1 Pro, Gemini 2.5 Pro, Claude Opus 4.5, Claude 3.7 Sonnet, Gemma 3 12B, InternVL3-8B, and Qwen3-VL-8B. 100 image families, 300 evaluation cells each.
Neil Tripathi tweet media
English
15
24
114
12.1K
RasG
RasG@Reportooor·
@Coredao_Org privacy features teased recently could tip the institutional scale, demand from regulated capital gonna hit hard once they drop these products🌊
English
0
0
1
1.4K
Core DAO 🔶
Core DAO 🔶@Coredao_Org·
Safe Bitcoin yield. 24/7 liquidity. Institutional-grade custody. The infrastructure is ready. The products are coming. 🔶
English
94
193
579
28.7K
RasG retweetledi
CoinMarketCap
CoinMarketCap@CoinMarketCap·
⚡ RECAP: The meme coin market just logged its first real bounce in weeks, with large caps rallying and a fresh AI meta lifting new launches. Bottom signal or dead cat bounce? Read on for the details this week. 👇🧵 1/7
CoinMarketCap tweet media
English
72
51
194
311.8K
RasG retweetledi
Watcher.Guru
Watcher.Guru@WatcherGuru·
JUST IN: $8,700,000,000 worth of Bitcoin & Ethereum options expire today.
English
453
438
4.7K
465.6K