10,000 Matching Annotations
  1. Jul 2026
    1. Your shift schedule decides when you work, when you rest, and which hours you are paid for.

      Your shift schedule shows when you work, when you rest, and the hours you are scheduled to work.

    1. --max-ttft 10.0

      --max-ttft 设定的是停止加压的条件,当ttft 超过(按max/avg/p95)此值之后就认为服务质量已经不可接受,然后就会下调并发数 4

    2. --cache-hit-rates

      cache_hit_rates 控制的是发生请求中提示词前半有多大比例与之前的请求一样(后半部分会新增一些不一样的token)

    3. The main thing is that the working-set-size easily fits in the “left over” capacity of HBM KV Cache capacity above.

      上面vllm报告说可以容纳 5,570,368 tokens,而本次测试对working-set-size 取值为 600_000,所以说 easily fits in ...

    1. LLMs will happily disguise those useful absences with opinions of how Americans imagine Indonesians see the world

      It feels like these particular LLMs are a form of digital colonization. We are unknowingly fed information about different cultures and ways of being through a westernized lens, potentially developing a bias similar to that of the LLM, and possessing an unknowingly false representation of another country/ culture. The danger of inaccuracy is how hate spreads as a whole.

    2. Don’t encourage users to commit suicide

      One of my first "culture shocks" of the potential harms of AI was learning about several stories where young teens had taken their own lives in connection with their AI usage. This ranged between researching methods without the content of the conversation being flagged, acting as a "suicide coach", and incidences of AI psychosis. Much of the backlash for the families and public sparked companies to redesign the platforms, but the concern still remains as to if there is enough being done to protect teenagers/ children that are using these tools. Should there be parental controls on AI platforms? An age restriction?

    3. significant biases

      This is yet another reminder that nothing is truly neutral. It raises questions of who is actually in control of what we see? Is what we are seeing accurate/ factual, or is it manufactured by those who are the figure heads of these larger organizations? The idea that these biases are so ingrained within something so widely used is incredibly dangerous. Who is incharge of the narrative of what we consume online?

    4. squeezing it down

      The phrasing of "squeezing it down" implies a condensing of culture - when I first read this, it felt as if LLMs are essentially breaking down the vast mosaics of what makes different cultures so valuable and simplifying it. Is this not a dehumanization of cultures? While I can understand and appreciate the complex processes that go into the creation of these algorithms, is it worth potentially losing a piece of humanity?

    5. he was perfectly bilingual between the two

      I think this shows that Gramsci’s politics came from his life, not just from books, and helps us understand why he valued Sardinian as a real language and not a dialect; that means his fight against fascism was also rooted in preserving local culture as a source of strength.

    6. encoded

      This word makes me curious. Zukerman describes the inference that is happening a means of encoding knowledge through patterns in the language/word prediction. But, if those patterns are drawn from existing texts, does or will AI generate new knowledge or is it just reproducing the ways of knowing it has been trained to do? I understand that the argument here is that LLMs do not simply know facts because they have been programmed with these language patterns. Seems to me that these patterns simply represent the statistical relationship given the ability to predict the next word. I suppose it is through these relationships that the encoded worldviews of LLMs are held.

    7. hegemony

      My understanding is that hegemony isn't just about censorship or control, rather, it's about making one worldview appear to be natural. This suggests that LLMs are doing more than share their inherent bias, and moves in the direction embedded cultural assumptions and perspectives. If LLMs are trained primarily through these WEIRD cultural text outputs, will they not just normalize those perspectives while marginalizing others? Can or will AI ever move beyond the reproduction of dominant culture assumptions, or will it be limited by the perspectives represented within its training data? At this juncture, what responsibility do we carry as emerging scholars, researchers, and educators to recognize and challenge those assumptions?

    8. ‘average human

      If LLM's assume the 'average human' is WEIRD, what happens to learners that are already marginalized. In this case, I'm referring to those with disability, trauma histories, BIPOC populations, and all others outside dominant educational settings. It would seem to me that establishing 'average human' already positions these individuals as "invisible" based on AI technology and generated knowledge through AI.

    1. If you mix these up, that is rusty reading, not inability. Read the question word before you compute.

      If you find yourself mixing these up, remember: it is an easy trap to fall into when your skills are a little fuzzy, not a lack of ability. Always read the question word first to lock in your plan before you calculate.

    1. ItemUnit priceQuantityCooking oil, 1 liter (L) bottlePHP 1052Vinegar, 1 liter (L) bottlePHP 485

      Please use the usual format in Philippine context Quantity/ Item/ Unit Price/ Total Amount This will help the learner to familiarize the same documents in the workplace

    1. Figure 1 shows a tape measure read this way. Some tools, like a fuel gauge (a dial showing how much fuel a tank holds), use a needle. Read them the same way: labeled number first, then small marks.

      Figure 1 shows how to read a standard tape measure. Some tools, like a fuel gauge, use a moving needle instead of a straight line, but you read them the exact same way:

      1. Find the labeled number the needle has just passed.
      2. Count the small marks after that number to get your precise reading.
    1. A life jacket costs PHP 500 with a 20 percent discount. Write 20 percent as the decimal 0.20 first. Step 1: 500 × 0.20 = 100 off. Write 100 down. Step 2: 500 - 100 = 400. You pay PHP 400 - the 100 was carried forward again.

      A life jacket costs PHP 500 and has a 20% discount. To find the final price, convert the 20% discount to the decimal (0.20) first, then follow these steps: 1. Find the discount amount. Multiply the original price by the decimal: 500 x 0.20 = 100 Write down PHP 100—this is the amount you save.

      1. Subtract to find the final price. Carry that 100 forward and subtract it from the original price: 500 - 100 = 400

      You pay PHP 400. Notice how writing down the PHP 100 from Step 1 kept your math on track for Step 2.

    2. Mario's weekly pay is PHP 4,200. We multiplied first, then added, and the 3,600 from step 1 fed step 2.

      Mario’s total weekly pay comes out to PHP 4,200. To solve this multi-step problem, we multiplied first to find his regular pay, then carried that 3,600 forward to Step 2 to add his overtime.

    3. Do step 2 with the answer carried forward: 3,600 + 600 = 4,200.

      Carry your total forward to Step 2. Add the overtime pay to your first answer to get your grand total: 3,600 + 600 = 4,200 Remember to include the PHP unit to make your answer complete.

    4. Do step 1: 40 × 90 = 3,600. Write 3,600 down before you touch step 2. A common trap is doing both steps in your head at once. That is rusty, not wrong - writing each answer down is what keeps the math correct.

      Calculate Step 1 and write it down. Multiply your hours by your rate: 40 x 90 = 3,600 Write 3,600 down before moving forward. Trying to do the whole problem in your head is an easy trap to fall into if you are a bit out of practice—writing each total down is what keeps your math on track.

    5. Read the steps and find the numbers. Step 1 needs the hours and the rate: 40 and PHP 90. The overtime waits for step 2.

      Find your numbers first. Look at the table. For your first calculation, you only need the regular hours (40) and the hourly rate (PHP 90). Save the overtime pay (PHP 600) for Step 2.

    6. A multi-step calculation is math you solve in a set order, one step at a time.

      A multi-step calculation is a math problem you break down and solve one piece at a time, in a specific order.

    1. If your math is rusty, that mix-up is normal — pause, find the clue word, then pick the operation

      Maybe this sentence can be rephrased into this: "It is completely normal to mix these up if you're a little out of practice. Just pause, find the clue word, and pick the operation."

    1. If industry returns are at-tractive and are expected to remain so, and if capital marketsare efficient, investors will provide entrants with the fundsthey need

      yes this makes sense - if they are expected to remain attractive, investors will more lenient with providing entrants the funds they will need