shetty (@thestoicccoder) header
shetty (@thestoicccoder) profile picture

a depressed PhD student

sf
Joined August 2024
Photo by shetty (@thestoicccoder)Photo by shetty (@thestoicccoder)Photo by shetty (@thestoicccoder)Photo by shetty (@thestoicccoder)Photo by shetty (@thestoicccoder)Photo by shetty (@thestoicccoder)Photo by shetty (@thestoicccoder)Photo by shetty (@thestoicccoder)Photo by shetty (@thestoicccoder)Photo by shetty (@thestoicccoder)
caminar de noche en una ciudad extranjera, debe ser top 3 sensaciones de esta vida
483
30,684
2,786
226,110
4,214,459
shetty retweeted
Unfair that CEOs who aren’t into Asian girls get fewer dating options from spy agencies. Where are the French/Russian/Iranian honeypots? Not one spy girl sent my way. Are they even trying? France is excused — their decision-makers aren’t into girls. Everyone else: try harder!
3,426
7,836
1,860
98,201
4,464,132
the pH of the vagina is so acidic it can completely dissolve $1.25T in market capitalization
1,336
20,777
1,603
220,676
4,396,379
shetty retweeted
here are some projects you can build to win your local college hackathon: - prove p=np - cure cancer - achieve human cloning - 10T param model - expense tracker app - reverse human ageing should be doable in 24 hours with ai.
32
14
4
317
8,681
shetty retweeted
i never cuda expected this
60
553
32
9,190
170,724
shetty retweeted
Never in my wildest dreams I thought this is going to be a trigger for ai bubble burst.
6
1
42
1,157
shetty retweeted
fable 5 when you ask it for the bedtime story your grandma used to tell you about enriching uranium
170
1,436
92
27,916
906,283
shetty retweeted
fable 5 reading the code written by opus 4.8 after going afk for 19 days
Readers added context they thought people might want to know
This video appears to be digitally altered. Here is an earlier version of this video, showing a different person instead of Haaland: x.com/kenw_2/status/…
47
232
15
6,570
659,024
shetty retweeted
Fable 5 is back.
4,230
10,962
5,014
89,880
17,363,139
Hermes Agent now reads the web up to 60x faster and 49x cheaper. Scraping backends pass clean content straight to the agent without redundant processing steps; large pages are saved locally and paged on demand so you get the same quality at a fraction of the time and cost.
202
391
94
6,176
812,028
shetty retweeted
yes that guy
Update: I've joined @AnthropicAI and taken leave from the university. Excited to work with many talented, mission-driven people on the defining technology of our time.
177
1,251
88
30,406
3,012,505
I knew it was going to be an insane nothingburger, because there's currently a soft ban on frontier capabilities but I genuinely don't understand why they didn't call it Sonnet 4.8 or Sonnet 4.9, because this artificially nerfed piece of shit is not worthy of the 5.0 naming
49
28
8
1,414
96,239
shetty retweeted
Anthropic hates Opensource AI because a model like GLM 5.2 being available for feee to download made their $1 Trillion valuation sound like a joke (which it is btw)
94
240
13
3,075
75,956
OpenAI achieved a much more significant breakthrough today. Sonnet 5 is an average release. But the fact that OpenAI, according to The Information, has managed to more than halve the inference costs of its current models through a new approach to inference optimization is absolutely groundbreaking. And when you also consider that they recently introduced their own inference chip with Broadcom, which is said to be faster and more efficient than the competition, I increasingly see OpenAI in an outstanding position. Today, at least, OpenAI emerges as the winner of the day.
Here is my first assessment of Sonnet 5: Sonnet 5 is better than Sonnet 4.6. Who would have thought? But jokes aside: Unfortunately, it is weaker than Opus 4.8 across all evals. Why they nevertheless labeled the latest Sonnet 5 iteration with a “5”, even though “4.8” would have been more fitting, is beyond me. Normally, major version jumps in particular signal a significant leap in capability. Be that as it may: Sonnet 5 is good, but worse than expected. Pricing has not changed; it is on the same level as its predecessor. Opus is still more expensive, but at the same time it also remains better. Overall, the release irritates me and leaves more questions than it answers. I cannot help but see Sonnet 5 as a release that stands in the context of Fable 5. There was no mention of Fable 5 at all, which surprises me a lot. I really would have expected us to get news about it at the same time. But nothing. Instead, we get an update to a new model series (“5”), but one that is not significant compared with the models we already have. As a result, there is a lingering aftertaste that Sonnet was released as something in between, perhaps also simply to release something at all and to stay part of the conversation, including in a positive sense. Why no Opus 5, when we know that Fable 5 already exists as a model that performs significantly better than 4.8, and when we can assume both that a better Opus exists internally and that it would not be difficult to update Opus to the new generation? Why “only” Sonnet 5? Because restraint is currently required. The major releases are currently being delayed across the board; they are still in discussions with regulators about how the truly powerful frontier releases can be carried out at all and under what conditions. In my view, the Sonnet 5 release has to be seen against this background. And as a result, at least for me, it was disappointing overall.
64
89
9
1,827
189,415
shetty retweeted
GLM 5.2 is ranking the highest on cost per session and everyone is raving about this model which means if cost/session is high it might actually be a sign that the model is useful
132
133
30
3,131
220,822
When you hit a wall in math, coding, or any hard skill, do not immediately conclude that you lack talent. Most walls are just prerequisite debt finally coming due. Go back, fill the gaps, make the basics automatic, and the wall often turns into a staircase.
44
599
37
4,775
367,514
> mythos is so good at cyber it can't be released also > mythos can't detect 20k fraudulent chinese accounts attacking it
344
970
103
18,087
515,016
I can't believe how many people believe Anthropic's propaganda about distillation attacks. You can read how GLM-5 was trained in the paper. Maybe Claude was distilled from Chinese models? I haven't seen a paper from Anthropic, and accusations are often confessions.
136
321
37
4,691
344,774
shetty retweeted
new inference optimization method by @deepseek_ai with an extremely detailed paper, draft model and framework to train them. results in production for dsv4 lead to +50% for throughput and latency (can go to ~80% for latency, crazy). full explanation of DSpark: it's about speculative decoding and the idea builds upon DFlash (fully parallel) and Eagle (fully sequential) to create a "semi-parallel" method that keeps the advantages of both the core equation you want to optimize is the "time to generate each token" which is: (time to draft + time to verify) / how many tokens are accepted the advantage of the parallel variant (DFlash) is that it's fast, but when you increase the number of tokens you draft, acceptance rate drops pretty fast (makes sense since there is no dependency on the previous token). fully sequential is nice but opposite issue: it's slower (you need a much smaller draft to get the same speed) but the autoregressive dependency means you can maintain good acceptance rate at a lot of tokens. since you have a much smaller draft head, the first token acceptance rate is often quite low idea of DSpark is to combine both: a "heavy" parallel head (you only do it once) and then a small sequential step to bias the logit distribution with information about the previous token. this biasing is done with a small markov head (only depends on t-1) they also get a confidence score out of the sequential head that allows them to adjust how many tokens they want to verify. verification can get expensive if the gpus are already at maximum utilization, so they use this confidence score to do some load balancing and predict the right number of tokens depending on gpu workload one small detail: i would have liked to see production numbers if they used DFlash or Eagle instead of MTP-1, but as always, huge work by deepseek and i'm expecting to see this method widely adopted
24
126
12
1,043
86,439
DAILY TECH NEWS ROUNDUP 🚨 Everything important that happened in tech during the last 24 hours: • India participated in the latest Pax Silica summit as the US led AI and semiconductor alliance expanded with new members including Italy. The coalition is becoming one of the most important efforts to secure AI chips, critical minerals, energy, and semiconductor supply chains across allied nations. • Apple raised prices across Mac, iPad, and HomePod products in India and several global markets after surging memory and storage chip costs. The move signals that AI driven semiconductor demand is now directly increasing consumer hardware prices. • Airwallex raised $320 million at an $11 billion valuation. The fintech plans to expand globally while investing further in AI capabilities, making it one of the day's largest confirmed funding announcements. • Anthropic accused Alibaba affiliated operators of conducting a massive model distillation campaign against Claude using tens of thousands of accounts. The allegation escalates the AI industry's battle over model extraction and intellectual property. • The White House asked OpenAI to limit the initial rollout of GPT-5.6 to a small group while additional safety evaluations are completed. The request highlights growing government involvement in frontier AI deployments before broad public release. • Micron said long term AI demand is fundamentally changing the memory chip business, joining Samsung and SK Hynix in pushing long duration supply agreements instead of traditional boom and bust cycles. The shift could reshape semiconductor pricing for years. • YouTube rolled out new Shorts features including faster playback and a Clear Screen viewing mode. The update strengthens YouTube's competition with TikTok by improving short form viewing and creator engagement. • Patronus AI announced a $50 million funding round to build large scale evaluation environments that stress test autonomous AI agents. Reliable AI evaluation is becoming a critical layer of enterprise AI deployment. • Australia is considering tougher enforcement mechanisms for its social media restrictions aimed at teenagers. The proposal could become one of the most consequential platform governance developments if adopted. • Google announced a new Finance app with expanded portfolio tracking and market intelligence features. The launch deepens Google's push into AI assisted personal finance experiences.
DAILY TECH NEWS ROUNDUP 🚨 Everything important that happened in tech during the last 24 hours: • 🤖 OpenAI revealed Jalapeño, its first custom AI inference chip developed with Broadcom. The chip is designed to power ChatGPT, Codex, the API, and future AI agents while reducing OpenAI's dependence on Nvidia GPUs. It marks a major step toward vertically integrated AI infrastructure. • Amazon announced a $13 billion investment in India to expand cloud and AI infrastructure. The move strengthens AWS capacity in one of the world's fastest growing technology markets and intensifies hyperscaler competition across Asia. • GitHub reported its strongest month ever for developer activity, driven by explosive adoption of AI coding tools and its new consumption-based Copilot pricing. The milestone highlights how AI-assisted software development is rapidly becoming the default workflow. • Micron delivered a stronger than expected outlook, sending shares sharply higher as demand for AI memory continues to surge. The results reinforce that high bandwidth memory remains one of the biggest beneficiaries of AI infrastructure spending. • 🚀 Cognition launched the Devin Ambassador Program. Selected developers receive free Devin Max access, credits, early feature access, direct communication with the Cognition team, and support for building local developer communities around Devin AI. • Anthropic hired Orange's AI chief to accelerate its European expansion. The appointment strengthens Anthropic's enterprise presence as competition among frontier AI companies continues to intensify. • Chinese AI company Z.ai said proceeds from its planned domestic listing will be used to pursue artificial general intelligence. The announcement underscores continued investment in frontier AI despite growing global competition. • 📈 Apple increased prices for several MacBook and iPad models across multiple markets as memory costs continue rising. The increases reflect higher DRAM and NAND prices driven by AI infrastructure demand rather than a broad price hike across Apple's entire product lineup. • Google released Apigee Emulator v2.0.1, patching ten security vulnerabilities affecting Netty and embedded Go components. Organizations using local API development environments are encouraged to upgrade. • GitLab shipped Community Edition and Enterprise Edition security updates fixing 13 vulnerabilities, including three rated high severity. The release is an important update for enterprises running self managed GitLab instances. • The Curl project released a new version resolving 18 security vulnerabilities, including fixes for a decades old issue uncovered during ongoing security reviews. The update affects one of the internet's most widely deployed open-source networking tools. • NIST opened an updated draft of its IoT cybersecurity guidance for public review. The proposal aims to strengthen baseline security requirements for internet connected devices and could influence future industry standards.
4
2
2
17
3,008