اکنون در دسترس! پژوهش تلگرام ۲۰۲۵ — مهمترین بینشهای سال 
Hitchhiking Inverse Log
رفتن به کانال در Telegram
AI stuff 🇺🇿➡️🇺🇸
نمایش بیشترکشور مشخص نشده استدسته بندی مشخص نشده است
2 910
مشترکین
اطلاعاتی وجود ندارد24 ساعت
-177 روز
-2530 روز
در حال بارگیری داده...
کانالهای مشابه
ابر برچسبها
اشارات ورودی و خروجی
---
---
---
---
---
---
جذب مشترکین
ژوئن '26
ژوئن '26
+51
در 5 کانالها
مه '26
+130
در 8 کانالها
Get PRO
آوریل '26
+263
در 19 کانالها
Get PRO
مارس '26
+1 440
در 3 کانالها
Get PRO
فوریه '260
در 4 کانالها
Get PRO
ژانویه '260
در 10 کانالها
Get PRO
دسامبر '250
در 6 کانالها
Get PRO
نوامبر '250
در 9 کانالها
Get PRO
اکتبر '250
در 3 کانالها
Get PRO
سپتامبر '250
در 5 کانالها
Get PRO
اوت '250
در 2 کانالها
Get PRO
ژوئیه '250
در 3 کانالها
Get PRO
ژوئن '250
در 7 کانالها
Get PRO
مه '250
در 1 کانالها
Get PRO
آوریل '250
در 3 کانالها
Get PRO
مارس '250
در 11 کانالها
Get PRO
فوریه '250
در 1 کانالها
Get PRO
ژانویه '250
در 0 کانالها
Get PRO
دسامبر '24
+20
در 1 کانالها
Get PRO
نوامبر '24
+31
در 0 کانالها
Get PRO
اکتبر '24
+9
در 0 کانالها
Get PRO
سپتامبر '24
+251
در 3 کانالها
Get PRO
اوت '24
+228
در 0 کانالها
Get PRO
ژوئیه '240
در 0 کانالها
Get PRO
ژوئن '240
در 0 کانالها
Get PRO
مه '240
در 1 کانالها
Get PRO
آوریل '240
در 0 کانالها
Get PRO
مارس '240
در 2 کانالها
Get PRO
فوریه '240
در 0 کانالها
Get PRO
ژانویه '240
در 0 کانالها
Get PRO
دسامبر '230
در 0 کانالها
Get PRO
نوامبر '230
در 0 کانالها
Get PRO
اکتبر '230
در 0 کانالها
Get PRO
سپتامبر '230
در 0 کانالها
Get PRO
اوت '230
در 0 کانالها
Get PRO
ژوئیه '230
در 0 کانالها
Get PRO
ژوئن '230
در 0 کانالها
Get PRO
مه '230
در 0 کانالها
Get PRO
آوریل '230
در 0 کانالها
Get PRO
مارس '230
در 0 کانالها
Get PRO
فوریه '230
در 0 کانالها
Get PRO
ژانویه '230
در 0 کانالها
Get PRO
دسامبر '22
+1
در 0 کانالها
Get PRO
نوامبر '22
+7
در 0 کانالها
Get PRO
اکتبر '22
+15
در 0 کانالها
Get PRO
سپتامبر '22
+8
در 0 کانالها
Get PRO
اوت '22
+9
در 0 کانالها
Get PRO
ژوئیه '22
+42
در 0 کانالها
Get PRO
ژوئن '22
+94
در 0 کانالها
Get PRO
مه '22
+25
در 0 کانالها
Get PRO
آوریل '22
+15
در 0 کانالها
Get PRO
مارس '22
+49
در 0 کانالها
Get PRO
فوریه '22
+5
در 0 کانالها
Get PRO
ژانویه '22
+44
در 0 کانالها
Get PRO
دسامبر '21
+23
در 0 کانالها
Get PRO
نوامبر '21
+36
در 0 کانالها
Get PRO
اکتبر '21
+146
در 0 کانالها
Get PRO
سپتامبر '21
+440
در 0 کانالها
Get PRO
اوت '21
+244
در 0 کانالها
| تاریخ | رشد مشترکین | اشارات | کانالها | |
| 29 ژوئن | +3 | |||
| 28 ژوئن | +2 | |||
| 27 ژوئن | +1 | |||
| 26 ژوئن | +1 | |||
| 25 ژوئن | 0 | |||
| 24 ژوئن | 0 | |||
| 23 ژوئن | 0 | |||
| 22 ژوئن | 0 | |||
| 21 ژوئن | 0 | |||
| 20 ژوئن | 0 | |||
| 19 ژوئن | +3 | |||
| 18 ژوئن | 0 | |||
| 17 ژوئن | +3 | |||
| 16 ژوئن | +1 | |||
| 15 ژوئن | +1 | |||
| 14 ژوئن | +3 | |||
| 13 ژوئن | +7 | |||
| 12 ژوئن | +2 | |||
| 11 ژوئن | +1 | |||
| 10 ژوئن | +3 | |||
| 09 ژوئن | +3 | |||
| 08 ژوئن | +1 | |||
| 07 ژوئن | +1 | |||
| 06 ژوئن | 0 | |||
| 05 ژوئن | +5 | |||
| 04 ژوئن | +4 | |||
| 03 ژوئن | +3 | |||
| 02 ژوئن | +1 | |||
| 01 ژوئن | +2 |
پستهای کانال
TIL there is a movie about Escher
https://youtu.be/nTxay3lAv1M?is=M9OYWgmgA1Q9peyb
| 2 | ⚫️ vs ⚪️ nodir.io/simulation | 1 224 |
| 3 |  بدون متن... | 1 290 |
| 4 |  +1The Allen Institute building in Seattle has a notable decorated walls. Zoomed in, these are red 0s and 1s, blue arrows and white letters. | 1 428 |
| 5 | They say "it will never be chill again". Anyway, Anthropic is big enough that you'd hear important enough news from other sources, so there is diminishing value in me posting them here. I'm overdue for some changes. ✌🏼 | 2 443 |
| 6 |  https://www.anthropic.com/news/fable-mythos-access | 2 408 |
| 7 |  +1fable beat gpt 5.5 at math | 2 847 |
| 8 | Abstractions and Progressive Disclosure
One distinct feature of Mythos/Fable is that it has better abstract thinking. This has always bothered me in Opus-class models: they didn't reuse code enough: when there is an existing piece of code, and the model would need a second piece which is similar but not the same, it would just write it from scratch, without extracting the common patterns. This has improved significantly with Mythos-class: I routinely see code reuse and refactoring here and there.
It has a second interesting effect. Not only it started recognizing patterns in code, but also in speech, and started introducing new abstractions in the language, ie words/terms, and using them in the text often, essentially compressing info. This makes its communication dense, and often hard to understand because it is referring to terms you either probably not familiar with, or just have never seen before because the model invented it on the fly. It often looks like an alien language. Neuralese.
I often have to say things like "explain in plain english assuming i have no context. It also helps to tell the model to use Progressive Disclosure in its comms, ie so that it does not talk about details that the receiver didn't ask for yet, and so that the level of abstraction is more or less uniform in a given paragraph/sentence (no jumping from high to low, with a bunch of details in parenthesis) | 1 834 |
| 9 | AI has a bubble problem: in the smaller startups that don't add a lot of value. Illustration:
https://x.com/elaifresh/status/2064907397043458141?s=20 | 52 |
| 10 |  +3if you have less than 2y of work experience (and presumably in US), and the following seems interesting, then check this out
https://www.anthropic.com/claude-corps/fellow | 1 778 |
| 11 | Emily Chang meets Anthropic co-founders Dario and Daniela Amodei for a rare, in-depth discussion of the startup's origin story, its battles with the Pentagon and how the company says it intends to put safety first in the high-stakes AI race.
https://youtu.be/v1wZwxY3CMg?is=FfR-GwDUDhPY-Pr0 | 1 639 |
| 12 |  +9"it's just pre-ipo marketing" | 1 763 |
| 13 | Concerned about AI's impact on the society/jobs/meaning? This essay is not that long - consider taking the time to read it in the original instead of TLDRing
https://darioamodei.com/post/policy-on-the-ai-exponential | 2 150 |
| 14 | Cybersecurity and biology jobs
[disclaimer: everything i write here should be treated as someone's attempt to interpret a very ambiguous situation. it will certainly be wrong sometimes and especially details]
Cybersecurity capabilities are dual use: can be used for both defense and attack. It is generally irresponsible to release a model with strong cybersecurity capabilities to everyone because hackers would use it for malevolent purposes. This is why Mythos is available only to certified organizations. Similarly, biology capabilities could be used to make biological weapons.
Models with such capabilities shouldn't be widely available, so those jobs won't be automated as fast (i have ideas on how this could be done but i'd rather not speculate). | 1 667 |
| 15 |  multiple good points. certainly worth watching at least this clip | 2 469 |
| 16 | Recursive Self Improvement
sometimes i feel cassandra-cursed: on one hand i feel some kind of responsibility to share critical predictions about ai to reduce the surprise. otoh the most critical predictions are also most unbelievable. we all remember the "it's just a smart autocomplete" cope.
another contributor to disbelief is that frontier labs and most people live in different realities: access to the best models, unlimited quota and agi-pilled-environment leads to futuristic systems that shape perspectives of only people in the labs, while the outside world often judges LLMs based on old/weak models or $20 plan (and, frankly, lack of imagination). The future is already here - it's just not evenly distributed (c). Arguably labs are closer to the future, but their informed predictions are often perceived as marketing.
Something we all can agree on is cold undeniable facts. Concrete evidence "from the future" would be easier to accept, but it is highly confidential, so all i can share is news and vague "trust me bro" predictions. In the meantime, we are getting closer and closer to transformative ai, the exponential isn't going to be accommodating and so anthropic also increasingly feels the obligation to prepare the world for the upcoming changes.
today's blog post states some facts about Anthropic's internal acceleration, demonstrates the trajectory to recursive self improvement (RSI) and talks about what it would take to pause or slowdown AI development. i won't quote anything here. please read the original:
https://www.anthropic.com/institute/recursive-self-improvement | 4 303 |
| 17 | ☣ One thing AI leaders can agree completely, is that it shouldn't be easy to make bio weapons. Open letter to congress: https://x.com/AndrewCurran_/status/2062347797710709177?s=20 | 1 844 |
| 18 | Anthropic is now profitable ($4.8B in q1) and OpenAI is rumored to IPO in September.
In related news, Anthropic will be paying $15B/y to SpaceX for compute, while SpaceX's total revenue is currently $18B/y.
The exponential continues | 3 017 |
| 19 | It marks the first time that a prominent open problem, central to a subfield of mathematics, has been solved autonomously by AI
The best known lower bound had been essentially unchanged since Erdős’s original 1946 construction.
The proof came from a new general-purpose reasoning model, rather than from a system trained specifically for mathematics, scaffolded to search through proof strategies, or targeted at the unit distance problem in particular.
the key ingredients of the construction come from a very different part of mathematics known as algebraic number theory, which studies concepts like factorization in extensions of the integers known as algebraic number fields.
https://openai.com/index/model-disproves-discrete-geometry-conjecture/ | 2 694 |
| 20 | and now second SpaceX data center (Colossus 2) will be used by anthropic
more compute ➡️ higher quotas
https://x.com/nottombrown/status/2057194829986300375?s=20 | 3 150 |
