Connect with us

Uncategorized

The DeepSeek-R1 Effect and Web3-AI

Published

5 месяцев ago

on

The artificial intelligence (AI) world was taken by storm a few days ago with the release of DeepSeek-R1, an open-source reasoning model that matches the performance of top foundation models while claiming to have been built using a remarkably low training budget and novel post-training techniques. The release of DeepSeek-R1 not only challenged the conventional wisdom surrounding the scaling laws of foundation models – which traditionally favor massive training budgets – but did so in the most active area of research in the field: reasoning.

The open-weights (as opposed to open-source) nature of the release made the model readily accessible to the AI community, leading to a surge of clones within hours. Moreover, DeepSeek-R1 left its mark on the ongoing AI race between China and the United States, reinforcing what has been increasingly evident: Chinese models are of exceptionally high quality and fully capable of driving innovation with original ideas.

Unlike most advancements in generative AI, which seem to widen the gap between Web2 and Web3 in the realm of foundation models, the release of DeepSeek-R1 carries real implications and presents intriguing opportunities for Web3-AI. To assess these, we must first take a closer look at DeepSeek-R1’s key innovations and differentiators.

Inside DeepSeek-R1

DeepSeek-R1 was the result of introducing incremental innovations into a well-established pretraining framework for foundation models. In broad terms, DeepSeek-R1 follows the same training methodology as most high-profile foundation models. This approach consists of three key steps:

Pretraining: The model is initially pretrained to predict the next word using massive amounts of unlabeled data.

Supervised Fine-Tuning (SFT): This step optimizes the model in two critical areas: following instructions and answering questions.

Alignment with Human Preferences: A final fine-tuning phase is conducted to align the model’s responses with human preferences.

Most major foundation models – including those developed by OpenAI, Google, and Anthropic – adhere to this same general process. At a high level, DeepSeek-R1’s training procedure does not appear significantly different. ButHowever, rather than pretraining a base model from scratch, R1 leveraged the base model of its predecessor, DeepSeek-v3-base, which boasts an impressive 617 billion parameters.

In essence, DeepSeek-R1 is the result of applying SFT to DeepSeek-v3-base with a large-scale reasoning dataset. The real innovation lies in the construction of these reasoning datasets, which are notoriously difficult to build.

First Step: DeepSeek-R1-Zero

One of the most important aspects of DeepSeek-R1 is that the process did not produce just a single model but two. Perhaps the most significant innovation of DeepSeek-R1 was the creation of an intermediate model called R1-Zero, which is specialized in reasoning tasks. This model was trained almost entirely using reinforcement learning, with minimal reliance on labeled data.

Reinforcement learning is a technique in which a model is rewarded for generating correct answers, enabling it to generalize knowledge over time.

R1-Zero is quite impressive, as it was able to match GPT-o1 in reasoning tasks. However, the model struggled with more general tasks such as question-answering and readability. That said, the purpose of R1-Zero was never to create a generalist model but rather to demonstrate it is possible to achieve state-of-the-art reasoning capabilities using reinforcement learning alone – even if the model does not perform well in other areas.

Second-Step: DeepSeek-R1

DeepSeek-R1 was designed to be a general-purpose model that excels at reasoning, meaning it needed to outperform R1-Zero. To achieve this, DeepSeek started once again with its v3 model, but this time, it fine-tuned it on a small reasoning dataset.

As mentioned earlier, reasoning datasets are difficult to produce. This is where R1-Zero played a crucial role. The intermediate model was used to generate a synthetic reasoning dataset, which was then used to fine-tune DeepSeek v3. This process resulted in another intermediate reasoning model, which was subsequently put through an extensive reinforcement learning phase using a dataset of 600,000 samples, also generated by R1-Zero. The final outcome of this process was DeepSeek-R1.

While I have omitted several technical details of the R1 pretraining process, here are the two main takeaways:

R1-Zero demonstrated that it is possible to develop sophisticated reasoning capabilities using basic reinforcement learning. Although R1-Zero was not a strong generalist model, it successfully generated the reasoning data necessary for R1.

R1 expanded the traditional pretraining pipeline used by most foundation models by incorporating R1-Zero into the process. Additionally, it leveraged a significant amount of synthetic reasoning data generated by R1-Zero.

As a result, DeepSeek-R1 emerged as a model that matched the reasoning capabilities of GPT-o1 while being built using a simpler and likely significantly cheaper pretraining process.

Everyone agrees that R1 marks an important milestone in the history of generative AI, one that is likely to reshape the way foundation models are developed. When it comes to Web3, it will be interesting to explore how R1 influences the evolving landscape of Web3-AI.

DeepSeek-R1 and Web3-AI

Until now, Web3 has struggled to establish compelling use cases that clearly add value to the creation and utilization of foundation models. To some extent, the traditional workflow for pretraining foundation models appears to be the antithesis of Web3 architectures. However, despite being in its early stages, the release of DeepSeek-R1 has highlighted several opportunities that could naturally align with Web3-AI architectures.

1) Reinforcement Learning Fine-Tuning Networks

R1-Zero demonstrated that it is possible to develop reasoning models using pure reinforcement learning. From a computational standpoint, reinforcement learning is highly parallelizable, making it well-suited for decentralized networks. Imagine a Web3 network where nodes are compensated for fine-tuning a model on reinforcement learning tasks, each applying different strategies. This approach is far more feasible than other pretraining paradigms that require complex GPU topologies and centralized infrastructure.

2) Synthetic Reasoning Dataset Generation

Another key contribution of DeepSeek-R1 was showcasing the importance of synthetically generated reasoning datasets for cognitive tasks. This process is also well-suited for a decentralized network, where nodes execute dataset generation jobs and are compensated as these datasets are used for pretraining or fine-tuning foundation models. Since this data is synthetically generated, the entire network can be fully automated without human intervention, making it an ideal fit for Web3 architectures.

3) Decentralized Inference for Small Distilled Reasoning Models

DeepSeek-R1 is a massive model with 671 billion parameters. However, almost immediately after its release, a wave of distilled reasoning models emerged, ranging from 1.5 to 70 billion parameters. These smaller models are significantly more practical for inference in decentralized networks. For example, a 1.5B–2B distilled R1 model could be embedded in a DeFi protocol or deployed within nodes of a DePIN network. More simply, we are likely to see the rise of cost-effective reasoning inference endpoints powered by decentralized compute networks. Reasoning is one domain where the performance gap between small and large models is narrowing, creating a unique opportunity for Web3 to efficiently leverage these distilled models in decentralized inference settings.

4) Reasoning Data Provenance

One of the defining features of reasoning models is their ability to generate reasoning traces for a given task. DeepSeek-R1 makes these traces available as part of its inference output, reinforcing the importance of provenance and traceability for reasoning tasks. The internet today primarily operates on outputs, with little visibility into the intermediate steps that lead to those results. Web3 presents an opportunity to track and verify each reasoning step, potentially creating a «new internet of reasoning» where transparency and verifiability become the norm.

Web3-AI Has a Chance in the Post-R1 Reasoning Era

The release of DeepSeek-R1 has marked a turning point in the evolution of generative AI. By combining clever innovations with established pretraining paradigms, it has challenged traditional AI workflows and opened a new era in reasoning-focused AI. Unlike many previous foundation models, DeepSeek-R1 introduces elements that bring generative AI closer to Web3.

Key aspects of R1 – synthetic reasoning datasets, more parallelizable training and the growing need for traceability – align naturally with Web3 principles. While Web3-AI has struggled to gain meaningful traction, this new post-R1 reasoning era may present the best opportunity yet for Web3 to play a more significant role in the future of AI.

Related Topics:
Continue Reading
Click to comment

Leave a Reply

Ваш адрес email не будет опубликован. Обязательные поля помечены *

Uncategorized

Bitcoin Network Hashrate Declined in June as Miners Reacted to Recent Heatwave: JPMorgan

Published

2 часа ago

on

1 июля, 2025

By

The Bitcoin (BTC) network monthly average hashrate fell about 3% in June, Wall Street bank JPMorgan (JPM) said in a research report Tuesday.

The hashrate refers to the total combined computational power used to mine and process transactions on a proof-of-work blockchain, and is a proxy for competition in the industry and mining difficulty. It is measured in exahashes per second (EH/s).

«Our sense is the decline was driven by seasonal weather-related curtailment in the U.S., and note that Cipher, IREN and Riot alone operate >80 EH/s in Texas,» analysts Reginald Smith and Charles Pearce wrote.

Bitcoin mining profitability continues to improve. The bank’s analysts estimated that miners earned an average of $55,300 per EH/s in daily block reward revenue last month, a 7% increase from April.

Daily block reward gross profit rose 13% month-on-month to the highest level since January, the analysts noted.

The total market cap of the 13 U.S.-listed bitcoin miners the bank follows rose 23%, or around $5.3 billion, from the previous month, the report said.

Operators with high-performance computing (HPC) exposure outperformed pure-play miners due to speculation of a deal between Core Scientific (CORZ) and CoreWeave (CRWV).

IREN (IREN) outperformed the group with a 67% gain, while Bitfarms (BITF) was the worst performer with a 19% decline, the report added.

Read more: U.S.-Listed Bitcoin Miners’ Share of Network Hashrate Hit Record High in June: JPMorgan

Continue Reading

Uncategorized

Crypto Daybook Americas: Bitcoin Posts Record Monthly Close, but Euro Steals the Show

Published

2 часа ago

on

1 июля, 2025

By

By Omkar Godbole (All times ET unless indicated otherwise)

Bitcoin (BTC) ended June above $107,000 at a record monthly close. Still, the largest crypto’s 2.5% monthly gain failed to match the euro’s advance against the dollar, the most liquid FX pair in the world.

The eurozone currency rose nearly 4% against the greenback last month, hitting its highest since September 2021. That prompted some traders to switch to euro-pegged stablecoins, resulting in a notable increase in their market values.

The euro’s momentum highlights the continued broad-based decline in the U.S. currency, which means financial conditions will likely remain easy, even though the weakness hasn’t done much to lift a directionless BTC.

The prolonged range play has been widely attributed to selling by wallets with a history of holdings coins for over a year. The profit-taking continued on Monday, with on-chain realized gains hitting the $2.4 billion mark.

BTC was recently trading 0.6% lower over 24 hours at $106,500. Other tokens, including XRP, DOGE, SOL and ETH, followed suit, while BCH, ALGO, and PAXG stood out.

Some analysts called for patience in the wake of continued institutional adoption. On Monday, Germany’s savings bank network said it will enable crypto trading for clients within a year. Strategy disclosed another major BTC purchase last week, acquiring $531 million worth of BTC.

«While short-term momentum has faded, medium-term signals remain bullish, especially with corporate treasuries accelerating their accumulation pace. We slightly reduced exposure to protect capital but remain constructive — especially on altcoins with room to catch up,» Valentin Fournier, lead research analyst at BRN.

That said, the third quarter has historically been bitcoin’s weakest. Moreover, liquidity tends to be weaker as well due to summer holidays, which raises the likelihood of exaggerated price moves. Remember the yen-led crash in BTC from $70,000 to $50,000 in late July to early August last year? This calls for caution even as analysts maintain the long-term constructive outlook.

In other news, American Bitcoin, a crypto firm backed by Eric Trump, raised $220 million to buy bitcoin and mining equipment. An FTX creditor posted on X that claims under $50,000 received 120% payouts in February and May 2025.

Bloomberg ETF analysts James Seyffart and Eric Balchunas said that there’s a 95% chance the U.S. SEC will approve spot ETFs for LTC and XRP this year.

In traditional markets, analysts awaited the Fed Chairman Jerome Powell’s speech later Tuesday and Friday’s nonfarm payrolls. Stay alert!

What to Watch

  • Crypto
    • July 2: Shares of the REX-Osprey Solana Staking ETF (tSSK) are expected to begin trading on the Cboe BZX Exchange, making this the first U.S.-listed ETF to combine SOL price exposure with on-chain staking rewards.
  • Macro
    • Day 2 of 3: ECB Forum on Central Banking (Sintra, Portugal)
    • July 1, 9 a.m.: S&P Global releases June Brazil data on manufacturing and services activity.
      • Manufacturing PMI Prev. 49.4
    • July 1, 9:30 a.m.: “High Level Policy Panel” discussion chaired by Fed Chair Jerome H. Powell at the ECB Forum on Central Banking in Sintra, Portugal. Livestream link.
    • July 1, 9:45 a.m.: S&P Global releases (final) June U.S. data on manufacturing and services activity.
      • Manufacturing PMI Est. 52 vs. Prev. 52
    • July 1, 10 a.m.: The Institute for Supply Management (ISM) releases June U.S. services sector data.
      • Manufacturing PMI Est. Est. 48.8 vs. Prev. 48.5
    • July 1, 10 a.m.: The U.S. Bureau of Labor Statistics releases April U.S. labor market data (i.e. the JOLTS report).
      • Job Openings Est. 7.3M vs. Prev. 7.391M
      • Job Quits Prev. 3.194M
    • July 2, 9:30 a.m.: S&P Global releases June Canada data on manufacturing and services activity.
      • Manufacturing PMI Prev. 46.1
    • July 3, 8:30 a.m.: The U.S. Bureau of Labor Statistics releases June employment data.
      • Non Farm Payrolls Est. 110K vs. Prev. 139K
      • Unemployment Rate Est. 4.3% vs. Prev. 4.2%
      • Government Payrolls Prev. -1K
      • Manufacturing Payrolls Est. -6K vs. Prev. -8K
    • July 3, 8:30 a.m.: The U.S. Department of Labor releases unemployment insurance data for the week ended June 28.
      • Initial Jobless Claims Est. 240K vs. Prev. 236K
      • Continuing Jobless Claims Est. 1960K vs. Prev. 1974K
    • July 3, 9 a.m.: S&P Global releases June Brazil data on manufacturing and services activity.
      • Composite PMI Prev. 49.1
      • Services PMI Prev. 49.6
    • July 3, 9:45 a.m.: S&P Global releases (final) June U.S. data on manufacturing and services activity.
      • Composite PMI Est. 52.8 vs. Prev. 53
      • Services PMI Est. 53.1 vs. Prev. 53.7
    • July 3, 10 a.m.: The Institute for Supply Management (ISM) releases June U.S. services sector data.
      • Services PMI Est. 50.5 vs. Prev. 49.9
  • Earnings (Estimates based on FactSet data)
    • None in the near future.

Token Events

  • Governance votes & calls
    • GnosisDAO is voting on renewing its partnership with Nethermind for Gnosis Chain maintenance and development, proposing 750,000 DAI funding for the first year from June, with 4% annual increases. Voting ends July 2.
    • Radiant DAO is voting on potentially compensating users whose wallets were drained via unlimited token approvals in the October 2024 hack. If passed, a follow-up plan would outline stablecoin conversions, claim contracts on Arbitrum, and phased repayments. Voting ends July 2.
    • Arbitrum DAO is voting on lowering the constitutional quorum threshold from 5% to 4.5% of votable tokens. This aims to match decreased voter participation and help well-supported proposals pass more easily, without affecting non-constitutional proposals, which remain at a 3% quorum. Voting ends July 4.
    • Polkadot Community is voting on launching a non-custodial Polkadot branded payment card to “to bridge the gap between digital assets in the Polkadot ecosystem and everyday spending.” Voting ends July 9.
  • Unlocks
    • July 1: Sui (SUI) to unlock 1.3% of its circulating supply worth $122.75 million.
    • July 2: Ethena (ENA) to unlock 0.67% of its circulating supply worth $10.59 million.
    • July 11: Immutable (IMX) to unlock 1.31% of its circulating supply worth $10.65 million.
    • July 12: Aptos (APT) to unlock 1.76% of its circulating supply worth $52.7 million.
    • July 15: Starknet (STRK) to unlock 3.79% of its circulating supply worth $14.42 million.
    • July 15: Sei (SEI) to unlock 1% of its circulating supply worth $15.73 million.
    • July 16: Arbitrum (ARB) to unlock 1.87% of its circulating supply worth $30.33 million.
  • Token Launches
    • July 4: Biswap (BSW), Stella (ALPHA), Komodo (KMD), LeverFi (LEVER), and LTO Network (LTO) to be delisted from Binance.

Conferences

The CoinDesk Policy & Regulation conference (formerly known as State of Crypto) is a one-day boutique event held in Washington on Sept. 10 that allows general counsels, compliance officers and regulatory executives to meet with public officials responsible for crypto legislation and regulatory oversight. Space is limited. Use code CDB10 for 10% off your registration through July 17.

Token Talk

By Francisco Rodrigues

  • While all eyes were on the introduction of Robinhood’s tokenized stocks and on Kraken and Bybit’s xStocks debut, a layer-2 network built to streamline DeFi quietly launched its mainnet yesterday.
  • Katana’s mainnet went live after it saw pre-deposits near $250 million, according to DeFiLlama data. The blockchain is backed by GSR and Polygon Labs.
  • The non-profit Katana Foundation says the chain attacks three chronic pain points: thin liquidity, erratic yields and capital flight. It does so by folding yield generation into the base layer.
  • When users bridge USDC, ETH, WBTC, AUSD or USDT, Katana’s VaultBridge pushes those funds into lending pools such as those on Morpho and Sushi, then sends the earnings back to depositors and app builders.
  • A separate mechanism called chain-owned liquidity captures transaction fees to bankroll the network over time.
  • Katana is distributing its native token, KAT, through liquidity mining. KAT is non-transferable for now, but the team expects an exchange listing by next year. Holders will be able to lock tokens for vKAT and share in staking rewards.

Derivatives Positioning

  • Perpetual funding rates for most tokens major tokens, including BTC and ETH, held marginally positive. XRP led with near 10% rates while XLM and ADA showed bias for shorts with sub-zero readings.
  • On the CME, BTC and ETH futures basis remained locked in the annualized 7% to 10% range.
  • On Deribit, risk reversals out to August-end expiry showed a bias for protective puts, with subsequent tenors showing a mild bias for calls. In ETH’s case, bearishness in the short-term tenors was more pronounced.
  • Block flows over the OTC desk Paradigm showed demand for the September expiry BTC $180K call option.

Market Movements

  • BTC is down 0.91% from 4 p.m. ET Monday at $106,629.81 (24hrs: -0.96%)
  • ETH is down 1.81% at $2,458.53 (24hrs: +0.15%)
  • CoinDesk 20 is down 2.37% at 3,010.77 (24hrs: -0.17%)
  • Ether CESR Composite Staking Rate is up 7 bps at 2.96%
  • BTC funding rate is at 0.0048% (5.3042% annualized) on Binance

CoinDesk 20 members’ performance

  • DXY is down 0.47% at 96.42
  • Gold futures are up 1.49% at $3,357.10
  • Silver futures are up 1.81% at $36.50
  • Nikkei 225 closed down 1.24% at 39,986.33
  • Hang Seng closed down 0.87% at 24,072.28
  • FTSE is down 0.17% at 8,745.89
  • Euro Stoxx 50 is down 0.30% at 5,287.47
  • DJIA closed on Monday up 0.63% at 44,094.77
  • S&P 500 closed up 0.52% at 6,204.95
  • Nasdaq Composite closed up 0.47% at 20,369.73
  • S&P/TSX Composite closed up 0.62% at 26,857.11
  • S&P 40 Latin America closed up 1.41% at 2,694.58
  • U.S. 10-Year Treasury rate is down 2.9 bps at 4.197%
  • E-mini S&P 500 futures are down 0.26% at 6,237.50
  • E-mini Nasdaq-100 futures are down 0.33% at 22,817.75
  • E-mini Dow Jones Industrial Average Index are down 0.13% at 44,331.00

Bitcoin Stats

  • BTC Dominance: 65.34% (0.19%)
  • Ethereum to bitcoin ratio: 0.02307 (-0.6%)
  • Hashrate (seven-day moving average): 869 EH/s
  • Hashprice (spot): $57.97
  • Total Fees: 4.22 BTC / $455,433
  • CME Futures Open Interest: 147,470 BTC
  • BTC priced in gold: 32.2 oz
  • BTC vs gold market cap: 9.12%

Technical Analysis

BTC's daily chart. (TradingView/CoinDesk)

  • BTC fell 1% Monday, narrowly missing the bull flag breakout. The decline produced a bearish outside day candle, with a price range wider than the preceding day’s candle.
  • Bearish outside day candles appearing after notable price gains, as in BTC’s case, signal renewed bearish trends.

Crypto Equities

  • Strategy (MSTR): closed on Monday at $404.23 (+5.3%), -1.64% at $397.59 in pre-market
  • Coinbase Global (COIN): closed at $350.49 (-0.83%), -1.53% at $345.12
  • Circle (CRCL): closed at $181.29 (+0.48%), +1.98% at $184.88
  • Galaxy Digital (GLXY): closed at $21.90 (+9.66%), +3.47% at $22.66
  • MARA Holdings (MARA): closed at $15.68 (+4.32%), -1.85% at $15.39
  • Riot Platforms (RIOT): closed at $11.3 (+7.11%), -1.59% at $11.12
  • Core Scientific (CORZ): closed at $17.07 (+2.52%), -1.52% at $16.81
  • CleanSpark (CLSK): closed at $11.03 (+3.37%), -1.81% at $10.83
  • CoinShares Valkyrie Bitcoin Miners ETF (WGMI): closed at $22.74 (+4.74%)
  • Semler Scientific (SMLR): closed at $38.74 (+0.62%), +0.15% at $38.80
  • Exodus Movement (EXOD): closed at $28.83 (-3.42%), +1.14% at $29.16

ETF Flows

Spot BTC ETFs

  • Daily net flows: $102.1 million
  • Cumulative net flows: $48.95 billion
  • Total BTC holdings ~1.25 million

Spot ETH ETFs

  • Daily net flows: $31.8 million
  • Cumulative net flows: $4.23 billion
  • Total ETH holdings ~4.1 million

Source: Farside Investors

Overnight Flows

Top 20 digital assets’ prices and volumes

Chart of the Day

Stablecoin transaction volume in USD. (Artemis)

  • The dollar value of the total stablecoin transactions crossed above the $4 trillion mark in June, the most since January, according to data source Artemis.
  • The data shows that while BTC’s price didn’t do much in the month, adoption of stablecoin continued unabated.

While You Were Sleeping

In the Ether

Last week was good for Bitcoin, but equities crushed it.Robinhood is launching its own L2 chain and you’re not bullish enough! You can’t make this up:Anyone who campaigned on the PROMISE of REDUCING SPENDINGTrump says Elon Musk would go broke and return to South Africa without US government subsidies for his businesses

Continue Reading

Uncategorized

The Blockchain Group Raises $13M to Advance Bitcoin Treasury Vision

Published

2 часа ago

on

1 июля, 2025

By

The Blockchain Group (ALTBG), listed on Euronext Growth Paris has secured roughly 11 million euros ($13 million) in fresh funding as it doubles down on becoming Europe’s first bitcoin (BTC) treasury company.

This strategic move underscores the firm’s commitment to growing its bitcoin holdings relative to its share count, aiming to deliver long-term value to investors through exposure to digital assets.

Part of the fundraising included $1.18 million capital increase at 5.251 euro per share, completed under an “ATM-type” agreement with asset manager TOBAM.

In parallel, the company’s wholly owned Luxembourg subsidiary issued 10 million euros ($11.8 million) in convertible bonds, priced at 5.174 euro per share, reflecting a 30 percent premium over the June 27 closing price. TOBAM subscribed for 5 million euros while bitcoin pioneer Adam Back invested around 5 million euros.

The Blockchain Group currently hold 1,794 BTC, while the share price is up 1% on Tuesday.

Continue Reading

Trending

Copyright © 2017 Zox News Theme. Theme by MVP Themes, powered by WordPress.