EasyPost (YC S13) Is Hiring
1 by jstreebin | 0 comments on Hacker News.
Senin, 30 September 2024
New sheriff in town: How Mike Macdonald has shaken up the Seahawks post-Pete Carroll
Seattle is 3-0 to start the season. Macdonald has preached accountability and conditioning with fines for tardiness.
from www.espn.com - TOP https://ift.tt/1ODys5j
from www.espn.com - TOP https://ift.tt/1ODys5j
Minggu, 29 September 2024
Follow live: Vikings look to continue hot start, visit rival Packers
null
from www.espn.com - TOP https://ift.tt/8ZGIlx3
from www.espn.com - TOP https://ift.tt/8ZGIlx3
Sources: Cowboys' Lawrence to miss 4-8 weeks
Cowboys defensive end DeMarcus Lawrence will miss between four and eight weeks because of a Lisfranc injury to his right foot suffered in Thursday night's win over the Giants, sources told ESPN's Todd Archer.
from www.espn.com - TOP https://ift.tt/rqlj7TS
from www.espn.com - TOP https://ift.tt/rqlj7TS
Sabtu, 28 September 2024
New top story on Hacker News: Autossh – automatically restart SSH sessions and tunnels
Autossh – automatically restart SSH sessions and tunnels
12 by denysonique | 1 comments on Hacker News.
12 by denysonique | 1 comments on Hacker News.
Arsenal ratings: Saka, Trossard star in dramatic win vs. Leicester
Bukayo Saka and Leandro Trossard were stellar for Arsenal, which rallied late.
from www.espn.com - TOP https://ift.tt/N7dgchM
from www.espn.com - TOP https://ift.tt/N7dgchM
Jumat, 27 September 2024
Duke's Flagg tries to ease in amid high hopes
Duke freshman Cooper Flagg has done his best to avoid the hype and embrace the competition.
from www.espn.com - TOP https://ift.tt/N53Vkwg
from www.espn.com - TOP https://ift.tt/N53Vkwg
Christine Sinclair to retire at end of NWSL season
Portland Thorns forward Christine Sinclair, the leading goal scorer in international soccer history, will retire at the end of the 2024 National Women's Soccer League season, she announced on Friday.
from www.espn.com - TOP https://ift.tt/CiEHzDh
from www.espn.com - TOP https://ift.tt/CiEHzDh
Kamis, 26 September 2024
Solskjaer open to Man United return despite sack
Ole Gunnar Solskjaer has said he would "say yes every day of the week" if asked by Manchester United to return to Old Trafford as manager.
from www.espn.com - TOP https://ift.tt/f7Gko8q
from www.espn.com - TOP https://ift.tt/f7Gko8q
Rabu, 25 September 2024
Facts vs. Feelings: How to deal with fantasy surprises, good and bad
Liz Loza deals with seven players who have been pleasant or unpleasant surprises so far this season.
from www.espn.com - TOP https://ift.tt/gRfsikl
from www.espn.com - TOP https://ift.tt/gRfsikl
Selasa, 24 September 2024
Source: Saints Pro Bowler McCoy needs surgery
Saints Pro Bowl center Erik McCoy, who exited Sunday's loss after just three snaps, needs groin surgery and is expected to miss time, a source confirmed to ESPN.
from www.espn.com - TOP https://ift.tt/wusTf1i
from www.espn.com - TOP https://ift.tt/wusTf1i
Senin, 23 September 2024
UNC's Brown: 'Disappointed' I talked resignation
UNC coach Mack Brown said he was "disappointed" for effectively offering his resignation to the team if players believed he wasn't capable of leading the program.
from www.espn.com - TOP https://ift.tt/vbXitpP
from www.espn.com - TOP https://ift.tt/vbXitpP
And 1: Raptors join Nets in retiring Carter's 15
The Raptors will retire Vince Carter's No. 15 jersey on Nov. 2, the first of two jersey retirement ceremonies for Carter. He'll also be honored by the Nets in January.
from www.espn.com - TOP https://ift.tt/4TQGPef
from www.espn.com - TOP https://ift.tt/4TQGPef
'You can't be yourself': Is Verstappen being pushed out of F1?
After being penalised for swearing in a news conference, Max Verstappen is hinting at leaving F1. Has the crackdown on foul language gone too far?
from www.espn.com - TOP https://ift.tt/0mlzrPU
from www.espn.com - TOP https://ift.tt/0mlzrPU
Minggu, 22 September 2024
Betting tips for 'Sunday Night Football': Chiefs at Falcons
Thinking of betting Chiefs-Falcons on 'Sunday Night Football'? Here's what you need to know to help your chances.
from www.espn.com - TOP https://ift.tt/Gb1Umai
from www.espn.com - TOP https://ift.tt/Gb1Umai
Sabtu, 21 September 2024
Jags TE Engram out for Monday's game at Bills
Jacksonville Jaguars tight end Evan Engram will miss Monday's game against Buffalo with a hamstring injury.
from www.espn.com - TOP https://ift.tt/xAI5Mac
from www.espn.com - TOP https://ift.tt/xAI5Mac
Verstappen protests presser after FIA penalty
Max Verstappen refused to give full answers during a post-qualifying press conference at the Singapore Grand Prix in an apparent protest against the governing body's decision to penalise him for swearing during Thursday's media session.
from www.espn.com - TOP https://ift.tt/D7CGLV4
from www.espn.com - TOP https://ift.tt/D7CGLV4
Liverpool ratings: Diaz, Gravenberch pull the strings in victory over Bournemouth
Liverpool picked up their fourth win in five league matches under Arne Slot thanks to Luis Diaz and Ryan Gravenberch.
from www.espn.com - TOP https://ift.tt/lgybJ6f
from www.espn.com - TOP https://ift.tt/lgybJ6f
Jumat, 20 September 2024
New top story on Hacker News: Show HN: Put this touch sensor on a robot and learn super precise tasks
Show HN: Put this touch sensor on a robot and learn super precise tasks
38 by raunaqmb | 3 comments on Hacker News.
We just released a very excited touch sensor that finally simplifies touch sensing for robotics. Our most exciting result: Learned visuotactile policies for precise tasks like inserting USBs and credit card swiping, that work out-of-the-box when you replace skins! To the best of our knowledge, this has never been shown before with any existing tactile sensor. Why is this important? For the first time, you could now collect data and train models on one sensor and expect them to generalize to new copies of the sensor -- opening the door to the kind of large foundation models that have revolutionized vision and language reasoning. Would love to hear the community's questions, thoughts and comments!
38 by raunaqmb | 3 comments on Hacker News.
We just released a very excited touch sensor that finally simplifies touch sensing for robotics. Our most exciting result: Learned visuotactile policies for precise tasks like inserting USBs and credit card swiping, that work out-of-the-box when you replace skins! To the best of our knowledge, this has never been shown before with any existing tactile sensor. Why is this important? For the first time, you could now collect data and train models on one sensor and expect them to generalize to new copies of the sensor -- opening the door to the kind of large foundation models that have revolutionized vision and language reasoning. Would love to hear the community's questions, thoughts and comments!
Sources: Top NFL CB prospect Revel Jr. tears ACL
One of the top prospects at cornerback for the NFL draft, East Carolina's Shavon Revel Jr. tore an ACL in practice and is out for the season, sources told ESPN.
from www.espn.com - TOP https://ift.tt/9wbUEqe
from www.espn.com - TOP https://ift.tt/9wbUEqe
Kamis, 19 September 2024
Rabu, 18 September 2024
Bolts' Herbert has ankle injury but feeling better
Chargers quarterback Justin Herbert has a right ankle injury but is "feeling much better," coach Jim Harbaugh said Wednesday.
from www.espn.com - TOP https://ift.tt/K7ThxrL
from www.espn.com - TOP https://ift.tt/K7ThxrL
Selasa, 17 September 2024
Flyers and goalie prospect Kolosov in standoff
The Flyers are in a standoff with goalie prospect Alexei Kolosov.
from www.espn.com - TOP https://ift.tt/GC0Xp9N
from www.espn.com - TOP https://ift.tt/GC0Xp9N
Senin, 16 September 2024
New top story on Hacker News: Ask HN: What runs L4-related microkernels/hypervisors these days?
Ask HN: What runs L4-related microkernels/hypervisors these days?
10 by AlexWandell | 1 comments on Hacker News.
I've been learning about the L4 microkernel, and am thinking about doing something related to it for a research project. I'm especially curious about more recent examples of specific devices that run L4 variants (seL4, PikeOS, OKL4, etc.). I already found a few that use seL4, but to take OKL4 as an example, most of the specific devices I could find are from more than a decade ago, and I'm trying to find things from at least the last 5 or 6 years. I'm even more curious to find devices that use a form of L4 as a hypervisor. Has anyone here worked on a device that used an L4-related kernel or hypervisor? I know one major area they're used in is defense and full of NDAs, but hopefully some of the other industries they're used in (medical devices, automotive, IoT) are a little less restrictive. Thanks in advance!
10 by AlexWandell | 1 comments on Hacker News.
I've been learning about the L4 microkernel, and am thinking about doing something related to it for a research project. I'm especially curious about more recent examples of specific devices that run L4 variants (seL4, PikeOS, OKL4, etc.). I already found a few that use seL4, but to take OKL4 as an example, most of the specific devices I could find are from more than a decade ago, and I'm trying to find things from at least the last 5 or 6 years. I'm even more curious to find devices that use a form of L4 as a hypervisor. Has anyone here worked on a device that used an L4-related kernel or hypervisor? I know one major area they're used in is defense and full of NDAs, but hopefully some of the other industries they're used in (medical devices, automotive, IoT) are a little less restrictive. Thanks in advance!
Pens re-sign Crosby to 2-year, $17.4M extension
Penguins captain Sidney Crosby signed a two-year contract extension worth $17.4 million that runs through the 2026-27 season on Monday.
from www.espn.com - TOP https://ift.tt/1Cuvg3W
from www.espn.com - TOP https://ift.tt/1Cuvg3W
Minggu, 15 September 2024
No Rice, no Ødegaard, no problem: Arsenal top Spurs in derby
Arsenal were without their first-choice midfield in the North London derby, but that didn't stop them from going to Spurs and taking all three points.
from www.espn.com - TOP https://ift.tt/MIFNJW8
from www.espn.com - TOP https://ift.tt/MIFNJW8
Sabtu, 14 September 2024
Mets promote IF Acuna, brother of Braves star
Seeking to strengthen the bench for a potential postseason run, the Mets promoted Luisangel Acuna, the younger brother of Braves star Ronald Acuna Jr., from Triple-A Syracuse.
from www.espn.com - TOP https://ift.tt/M5DzfeT
from www.espn.com - TOP https://ift.tt/M5DzfeT
Liverpool stunned by Forest at Anfield for 1st loss
A superb second-half goal from substitute Callum Hudson-Odoi gave Nottingham Forest the biggest surprise win of the Premier League season so far as they beat Liverpool 1-0 at Anfield on Saturday.
from www.espn.com - TOP https://ift.tt/D3LuVeK
from www.espn.com - TOP https://ift.tt/D3LuVeK
Twins SS Correa activated, to return vs. Reds
The Twins activated All-Star shortstop Carlos Correa from the 10-day injured list ahead of Saturday's game against the Reds.
from www.espn.com - TOP https://ift.tt/FdEYev7
from www.espn.com - TOP https://ift.tt/FdEYev7
Slain Olympic runner Cheptegei buried in Uganda
Olympic runner Rebecca Cheptegei, who died after allegedly being doused in petrol and set alight by her former partner, was buried Saturday with full military honors in her native Uganda.
from www.espn.com - TOP https://ift.tt/d31vwMr
from www.espn.com - TOP https://ift.tt/d31vwMr
Jumat, 13 September 2024
Kellen Winslow II seeks reduced prison sentence
Former NFL tight end Kellen Winslow II is seeking a reduction of his prison sentence, citing new criminal justice laws in California.
from www.espn.com - TOP https://ift.tt/SLoq3bH
from www.espn.com - TOP https://ift.tt/SLoq3bH
Kamis, 12 September 2024
PSG told to pay Mbappe $61M in unpaid wages
The French league's legal commission has ordered Paris Saint-Germain to pay Kylian Mbappé the €55 million in unpaid wages that he claims he's entitled to, the league said Thursday.
from www.espn.com - TOP https://ift.tt/6kInpv1
from www.espn.com - TOP https://ift.tt/6kInpv1
Rabu, 11 September 2024
Rahm eyes World Tour to stay Ryder Cup-eligible
Jon Rahm said he intends to play in three DP World Tour events by the end of the season to remain eligible for next year's Ryder Cup.
from www.espn.com - TOP https://ift.tt/hsRPJpL
from www.espn.com - TOP https://ift.tt/hsRPJpL
Watson denies latest sexual assault allegations
Browns QB Deshaun Watson has denied allegations of sexual assault and battery that were levied in a civil lawsuit filed earlier this week in Texas.
from www.espn.com - TOP https://ift.tt/zZ2n8jQ
from www.espn.com - TOP https://ift.tt/zZ2n8jQ
Selasa, 10 September 2024
New top story on Hacker News: Show HN: GitOps Template for Kubernetes
Show HN: GitOps Template for Kubernetes
21 by pmig | 1 comments on Hacker News.
Hello HN, we’re Philip Louis from Glasskube ( https://ift.tt/GZPtzgE ). We are working on a package manager for Kubernetes to simplify the packaging of complex applications with multiple dependencies, ensuring they are installed and kept up-to-date across multiple Kubernetes clusters. Nowadays, it is best practice to use Git as a revision control system for your Kubernetes configurations. Update automation workflows like Renovate or Dependabot can create pull requests for new versions of Docker images and Helm charts, but ensuring these new package versions work is still a manual task. By using the central (or a private) Glasskube repository ( https://ift.tt/AThIqaF ) together with our Renovate integration ( https://ift.tt/KvoWH0V ), you can ensure that new package versions will run through our Minikube-based CI workflows before they get published—similar to how the Homebrew core tap works. We’ve just introduced readiness checks for manifest-based deployments and utilize the flux-helm-controller to wait for a Helm release to succeed. Dependencies are resolved by our package controller. These dependencies can either be cluster-scoped (installed in the recommended namespace, e.g., operators wird CRDs) or namespace-scoped components of a package (e.g., a database or Redis cache). In such cases, we will prefix resources with the dependent package name to ensure multiple packages can use the same dependencies without naming conflicts (we use Kustomize on a virtual filesystem for this). Glasskube packages can currently be Helm charts (from an OCI or Helm repository) or manifests, which are mostly built using Kustomize’s overlay approach. Since neither the overlay approach (using Kustomize) nor Helm’s limited templating functionality will help us and other Kubernetes users scale to more complex packages, we are considering creating a more programmatic approach to package creation, similar to Timoni. Currently, KCL is our frontrunner ( https://ift.tt/4oh2UDP ), as it already integrates well with the Kubernetes ecosystem. We would appreciate if you give our GitOps template a try. It also works work existing Kubernetes clusters if just want to use GitOps for some applications. Just make sure that the argocd and glasskube-system namespaces are not yet in use. See: https://ift.tt/2GwKX6e
21 by pmig | 1 comments on Hacker News.
Hello HN, we’re Philip Louis from Glasskube ( https://ift.tt/GZPtzgE ). We are working on a package manager for Kubernetes to simplify the packaging of complex applications with multiple dependencies, ensuring they are installed and kept up-to-date across multiple Kubernetes clusters. Nowadays, it is best practice to use Git as a revision control system for your Kubernetes configurations. Update automation workflows like Renovate or Dependabot can create pull requests for new versions of Docker images and Helm charts, but ensuring these new package versions work is still a manual task. By using the central (or a private) Glasskube repository ( https://ift.tt/AThIqaF ) together with our Renovate integration ( https://ift.tt/KvoWH0V ), you can ensure that new package versions will run through our Minikube-based CI workflows before they get published—similar to how the Homebrew core tap works. We’ve just introduced readiness checks for manifest-based deployments and utilize the flux-helm-controller to wait for a Helm release to succeed. Dependencies are resolved by our package controller. These dependencies can either be cluster-scoped (installed in the recommended namespace, e.g., operators wird CRDs) or namespace-scoped components of a package (e.g., a database or Redis cache). In such cases, we will prefix resources with the dependent package name to ensure multiple packages can use the same dependencies without naming conflicts (we use Kustomize on a virtual filesystem for this). Glasskube packages can currently be Helm charts (from an OCI or Helm repository) or manifests, which are mostly built using Kustomize’s overlay approach. Since neither the overlay approach (using Kustomize) nor Helm’s limited templating functionality will help us and other Kubernetes users scale to more complex packages, we are considering creating a more programmatic approach to package creation, similar to Timoni. Currently, KCL is our frontrunner ( https://ift.tt/4oh2UDP ), as it already integrates well with the Kubernetes ecosystem. We would appreciate if you give our GitOps template a try. It also works work existing Kubernetes clusters if just want to use GitOps for some applications. Just make sure that the argocd and glasskube-system namespaces are not yet in use. See: https://ift.tt/2GwKX6e
Steelers watching Wilson, prepping Fields to start
The Steelers will monitor Russell Wilson's progress this week as he recovers from a calf injury, but they're preparing for Justin Fields to start vs. the Broncos.
from www.espn.com - TOP https://ift.tt/w29OPUX
from www.espn.com - TOP https://ift.tt/w29OPUX
Senin, 09 September 2024
FSU tailback Lucas to miss remainder of season
Florida State will be without tailback Jaylin Lucas for the remainder of the season, head coach Mike Norvell said.
from www.espn.com - TOP https://ift.tt/h5VIvJz
from www.espn.com - TOP https://ift.tt/h5VIvJz
Minggu, 08 September 2024
NFL in his DNA: Kendrick Lamar will play SB LIX
Rap star and Grammy award winner Kendrick Lamar will headline the Super Bowl LIX halftime show in New Orleans.
from www.espn.com - TOP https://ift.tt/XMnqIsR
from www.espn.com - TOP https://ift.tt/XMnqIsR
With Prescott's mega-payday, his legacy rests on getting to the Super Bowl
After eight seasons as the starter, Dak Prescott needs a Super Bowl to cement his place among Cowboys greats.
from www.espn.com - TOP https://ift.tt/t4gfQRk
from www.espn.com - TOP https://ift.tt/t4gfQRk
Sabtu, 07 September 2024
Jets to start season without holdout DE Reddick
Jets coach Robert Saleh acknowledged that it "would be fair" to say the team will begin the season without Haason Reddick. The Pro Bowl defensive end has been holding out after being traded to New York this offseason.
from www.espn.com - TOP https://ift.tt/t0eOH6L
from www.espn.com - TOP https://ift.tt/t0eOH6L
23XI Racing skips deadline to sign charter deal
23XI Racing, the NASCAR team co-owned by Michael Jordan, said Saturday it skipped a deadline to sign a new charter agreement with the sanctioning body because "it did not have an opportunity to fairly bargain" for a new contract.
from www.espn.com - TOP https://ift.tt/qITcp15
from www.espn.com - TOP https://ift.tt/qITcp15
Jumat, 06 September 2024
The shark attack survivor going for glory at Paralympics
Ali Truwit swam for her life after she was attacked by a shark. Now she is representing Team USA in swimming at the Paralympics.
from www.espn.com - TOP https://ift.tt/blWTDes
from www.espn.com - TOP https://ift.tt/blWTDes
New top story on Hacker News: Show HN: Infinity – Realistic AI characters that can speak
Show HN: Infinity – Realistic AI characters that can speak
35 by lcolucci | 30 comments on Hacker News.
Hey HN, this is Lina from Infinity AI ( https://infinity.ai/ ). My co-founders and I have trained our own foundation video model focused on people. As far as we know, this is the first time someone has trained a video diffusion transformer that’s driven by audio input. This is cool because it allows for expressive, realistic-looking characters that actually speak. Here’s a blog with a bunch of examples: https://ift.tt/vD6Hsig If you want to try it out, you can either (1) go to https://ift.tt/SxVXfRh , or (2) post a comment in this thread describing a character and we’ll generate a video for you and reply with a link. For example: “Mona Lisa saying ‘what the heck are you smiling at?’”: https://bit.ly/3z8l1TM “A 3D pixar-style gnome with a pointy red hat reciting the Declaration of Independence”: https://bit.ly/3XzpTdS “Elon Musk singing Fly Me To The Moon by Sinatra”: https://bit.ly/47jyC7C Our tool at Infinity allows creators to type out a script with what they want their characters to say (and eventually, what they want their characters to do) and get a video out. We’ve trained for about 11 GPU years (~$500k) so far and our model recently started getting good results, so we wanted to share it here. We are still actively training. We had trouble creating videos of good characters with existing AI tools. Generative AI video models (like Runway and Luma) don’t allow characters to speak. And talking avatar companies (like HeyGen and Synthesia) just do lip syncing on top of the previously recorded videos. This means you often get facial expressions and gestures that don’t make sense with the audio, resulting in the “uncanny” look you can’t quite put your finger on. See blog. When we started Infinity, our V1 model took the lip syncing approach. In addition to mismatched gestures, this method had many limitations, including a finite library of actors (we had to fine-tune a model for each one with existing video footage) and an inability to animate imaginary characters. To address these limitations in V2, we decided to train an end-to-end video diffusion transformer model that takes in a single image, audio, and other conditioning signals and outputs video. We believe this end-to-end approach is the best way to capture the full complexity and nuances of human motion and emotion. One drawback of our approach is that the model is slow despite using rectified flow (2-4x speed up) and a 3D VAE embedding layer (2-5x speed up). Here are a few things the model does surprisingly well on: (1) it can handle multiple languages, (2) it has learned some physics (e.g. it generates earrings that dangle properly and infers a matching pair on the other ear), (3) it can animate diverse types of images (paintings, sculptures, etc) despite not being trained on those, and (4) it can handle singing. See blog. Here are some failure modes of the model: (1) it cannot handle animals (only humanoid images), (2) it often inserts hands into the frame (very annoying and distracting), (3) it’s not robust on cartoons, and (4) it can distort people’s identities (noticeable on well-known figures). See blog. Try the model here: https://ift.tt/SxVXfRh We’d love to hear what you think!
35 by lcolucci | 30 comments on Hacker News.
Hey HN, this is Lina from Infinity AI ( https://infinity.ai/ ). My co-founders and I have trained our own foundation video model focused on people. As far as we know, this is the first time someone has trained a video diffusion transformer that’s driven by audio input. This is cool because it allows for expressive, realistic-looking characters that actually speak. Here’s a blog with a bunch of examples: https://ift.tt/vD6Hsig If you want to try it out, you can either (1) go to https://ift.tt/SxVXfRh , or (2) post a comment in this thread describing a character and we’ll generate a video for you and reply with a link. For example: “Mona Lisa saying ‘what the heck are you smiling at?’”: https://bit.ly/3z8l1TM “A 3D pixar-style gnome with a pointy red hat reciting the Declaration of Independence”: https://bit.ly/3XzpTdS “Elon Musk singing Fly Me To The Moon by Sinatra”: https://bit.ly/47jyC7C Our tool at Infinity allows creators to type out a script with what they want their characters to say (and eventually, what they want their characters to do) and get a video out. We’ve trained for about 11 GPU years (~$500k) so far and our model recently started getting good results, so we wanted to share it here. We are still actively training. We had trouble creating videos of good characters with existing AI tools. Generative AI video models (like Runway and Luma) don’t allow characters to speak. And talking avatar companies (like HeyGen and Synthesia) just do lip syncing on top of the previously recorded videos. This means you often get facial expressions and gestures that don’t make sense with the audio, resulting in the “uncanny” look you can’t quite put your finger on. See blog. When we started Infinity, our V1 model took the lip syncing approach. In addition to mismatched gestures, this method had many limitations, including a finite library of actors (we had to fine-tune a model for each one with existing video footage) and an inability to animate imaginary characters. To address these limitations in V2, we decided to train an end-to-end video diffusion transformer model that takes in a single image, audio, and other conditioning signals and outputs video. We believe this end-to-end approach is the best way to capture the full complexity and nuances of human motion and emotion. One drawback of our approach is that the model is slow despite using rectified flow (2-4x speed up) and a 3D VAE embedding layer (2-5x speed up). Here are a few things the model does surprisingly well on: (1) it can handle multiple languages, (2) it has learned some physics (e.g. it generates earrings that dangle properly and infers a matching pair on the other ear), (3) it can animate diverse types of images (paintings, sculptures, etc) despite not being trained on those, and (4) it can handle singing. See blog. Here are some failure modes of the model: (1) it cannot handle animals (only humanoid images), (2) it often inserts hands into the frame (very annoying and distracting), (3) it’s not robust on cartoons, and (4) it can distort people’s identities (noticeable on well-known figures). See blog. Try the model here: https://ift.tt/SxVXfRh We’d love to hear what you think!
Kamis, 05 September 2024
Big 12's discussions to add UConn have stalled
The Big 12's conversations about potentially adding UConn as a member have stalled.
from www.espn.com - TOP https://ift.tt/dw5UcWP
from www.espn.com - TOP https://ift.tt/dw5UcWP
Rabu, 04 September 2024
New top story on Hacker News: Show HN: Graphiti – LLM-Powered Temporal Knowledge Graphs
Show HN: Graphiti – LLM-Powered Temporal Knowledge Graphs
17 by roseway4 | 2 comments on Hacker News.
Hey HN! We're Paul, Preston, and Daniel from Zep. We've just open-sourced Graphiti, a Python library for building temporal Knowledge Graphs using LLMs. Graphiti helps you create and query graphs that evolve over time. Knowledge Graphs have been explored extensively for information retrieval. What makes Graphiti unique is its ability to build a knowledge graph while handling changing relationships and maintaining historical context. At Zep, we build a memory layer for LLM applications. Developers use Zep to recall relevant user information from past conversations without including the entire chat history in a prompt. Accurate context is crucial for LLM applications. If an AI agent doesn't remember that you've changed jobs or confuses the chronology of events, its responses can be jarring or irrelevant, or worse, inaccurate. Before Graphiti, our approach to storing and retrieving user “memory” was, in effect, a specialized RAG pipeline. An LLM extracted “facts” from a user’s chat history. Semantic search, reranking, and other techniques then surfaced facts relevant to the current conversation back to a developer for inclusion in their prompt. We attempted to reconcile how new information may change our understanding of existing facts: Fact: “Kendra loves Adidas shoes” User message: “I’m so angry! My favorite Adidas shoes fell apart! Puma’s are my new favorite shoes!” Facts: - “Kendra used to love Adidas shoes but now prefers Puma.” - “Kendra’s Adidas shoes fell apart.” Unfortunately, this approach became problematic. Reconciling facts from increasingly complex conversations challenged even frontier LLMs such as gpt-4o. We saw incomplete facts, poor recall, and hallucinations. Our RAG search also failed at times to capture the nuanced relationships between facts, leading to irrelevant or contradictory information being retrieved. We tried fixing these issues with prompt optimization but saw diminishing returns on effort. We realized that a graph would help model a user’s complex world, potentially addressing these challenges. We were intrigued by Microsoft’s GraphRAG, which expanded on RAG text chunking with a graph to better model a document corpus. However, it didn't solve our core problem: GraphRAG is designed for static documents and doesn't natively handle temporality. So, we built Graphiti, which is designed from the ground up to handle constantly changing information, hybrid semantic and graph search, and scale: - Temporal Awareness: Tracks changes in facts and relationships over time. Graph edges include temporal metadata to record relationship lifecycles. - Episodic Processing: Ingests data as discrete episodes, maintaining data provenance and enabling incremental processing. - Hybrid Search: Semantic and BM25 full-text search, with the ability to rerank results by distance from a central node. - Scalable: Designed for large datasets, parallelizing LLM calls for batch processing while preserving event chronology. - Varied Sources: Ingests both unstructured text and structured data. Graphiti has significantly improved our ability to maintain accurate user context. It does a far better job of fact reconciliation over long, complex conversations. Node distance reranking, which places a user at the center of the graph, has also been a valuable tool. Quantitative data evaluation results may be a future ShowHN. Work is ongoing, including: 1. Improving support for faster and cheaper small language models. 2. Exploring fine-tuning to improve accuracy and reduce latency. 3. Adding new querying capabilities, including search over neighborhood (sub-graph) summaries. ## Getting Started Graphiti is open source and available on GitHub: https://ift.tt/pgTndrD . We'd love to hear your thoughts. Please also consider contributing!
17 by roseway4 | 2 comments on Hacker News.
Hey HN! We're Paul, Preston, and Daniel from Zep. We've just open-sourced Graphiti, a Python library for building temporal Knowledge Graphs using LLMs. Graphiti helps you create and query graphs that evolve over time. Knowledge Graphs have been explored extensively for information retrieval. What makes Graphiti unique is its ability to build a knowledge graph while handling changing relationships and maintaining historical context. At Zep, we build a memory layer for LLM applications. Developers use Zep to recall relevant user information from past conversations without including the entire chat history in a prompt. Accurate context is crucial for LLM applications. If an AI agent doesn't remember that you've changed jobs or confuses the chronology of events, its responses can be jarring or irrelevant, or worse, inaccurate. Before Graphiti, our approach to storing and retrieving user “memory” was, in effect, a specialized RAG pipeline. An LLM extracted “facts” from a user’s chat history. Semantic search, reranking, and other techniques then surfaced facts relevant to the current conversation back to a developer for inclusion in their prompt. We attempted to reconcile how new information may change our understanding of existing facts: Fact: “Kendra loves Adidas shoes” User message: “I’m so angry! My favorite Adidas shoes fell apart! Puma’s are my new favorite shoes!” Facts: - “Kendra used to love Adidas shoes but now prefers Puma.” - “Kendra’s Adidas shoes fell apart.” Unfortunately, this approach became problematic. Reconciling facts from increasingly complex conversations challenged even frontier LLMs such as gpt-4o. We saw incomplete facts, poor recall, and hallucinations. Our RAG search also failed at times to capture the nuanced relationships between facts, leading to irrelevant or contradictory information being retrieved. We tried fixing these issues with prompt optimization but saw diminishing returns on effort. We realized that a graph would help model a user’s complex world, potentially addressing these challenges. We were intrigued by Microsoft’s GraphRAG, which expanded on RAG text chunking with a graph to better model a document corpus. However, it didn't solve our core problem: GraphRAG is designed for static documents and doesn't natively handle temporality. So, we built Graphiti, which is designed from the ground up to handle constantly changing information, hybrid semantic and graph search, and scale: - Temporal Awareness: Tracks changes in facts and relationships over time. Graph edges include temporal metadata to record relationship lifecycles. - Episodic Processing: Ingests data as discrete episodes, maintaining data provenance and enabling incremental processing. - Hybrid Search: Semantic and BM25 full-text search, with the ability to rerank results by distance from a central node. - Scalable: Designed for large datasets, parallelizing LLM calls for batch processing while preserving event chronology. - Varied Sources: Ingests both unstructured text and structured data. Graphiti has significantly improved our ability to maintain accurate user context. It does a far better job of fact reconciliation over long, complex conversations. Node distance reranking, which places a user at the center of the graph, has also been a valuable tool. Quantitative data evaluation results may be a future ShowHN. Work is ongoing, including: 1. Improving support for faster and cheaper small language models. 2. Exploring fine-tuning to improve accuracy and reduce latency. 3. Adding new querying capabilities, including search over neighborhood (sub-graph) summaries. ## Getting Started Graphiti is open source and available on GitHub: https://ift.tt/pgTndrD . We'd love to hear your thoughts. Please also consider contributing!
Brazil beefs up security amid NFL player concerns
Sao Paulo has tightened security ahead of the first NFL game in South America after players voiced concerns about traveling to Latin America's largest country, the Brazilian state said Wednesday.
from www.espn.com - TOP https://ift.tt/iHUScGO
from www.espn.com - TOP https://ift.tt/iHUScGO
Bills' Hamlin to get first start since cardiac arrest
Bills safety Damar Hamlin, who suffered cardiac arrest in January 2023, will make his first start since then in Week 1, coach Sean McDermott said Wednesday.
from www.espn.com - TOP https://ift.tt/BsqQ5wD
from www.espn.com - TOP https://ift.tt/BsqQ5wD
Why Jannik Sinner's forehand is unique -- and so dangerous on the tennis court
The distinctive sound of Jannik Sinner's forehand can make a crowd gasp. And it's his not-so-secret tool in any match.
from www.espn.com - TOP https://ift.tt/oSKTWac
from www.espn.com - TOP https://ift.tt/oSKTWac
Selasa, 03 September 2024
Judge denies Rocha appeal to run as RFEF chief
A judge has denied a request from Spanish football federation (RFEF) president Pedro Rocha to grant a temporary injunction against his suspension, leaving him unable to continue in the position, barring an urgent appeal.
from www.espn.com - TOP https://ift.tt/RmST4iQ
from www.espn.com - TOP https://ift.tt/RmST4iQ
Senin, 02 September 2024
Fins' Hill, Waddle practice after summer injuries
After battling injuries and not playing during the preseason, Dolphins receivers Tyreek Hill and Jaylen Waddle returned to practice ahead of the opener vs. the Jaguars.
from www.espn.com - TOP https://ift.tt/Wj1rghm
from www.espn.com - TOP https://ift.tt/Wj1rghm
Minggu, 01 September 2024
Yankees 1B Rizzo activated after 11 weeks on IL
The Yankees reinstated infielder Anthony Rizzo from the 60-day injured list on Sunday. He had been working his way back after sustaining a right forearm fracture in a collision with Red Sox pitcher Brennan Bernardino on June 16.
from www.espn.com - TOP https://ift.tt/BStKFoi
from www.espn.com - TOP https://ift.tt/BStKFoi
'Pain of watching someone else': Juice Kiffin takes to social media after injury
Juice, Ole Miss head coach Lane Kiffin's dog, has retrieved the kickoff tee at Vaught Hemingway Stadium since last September.
from www.espn.com - TOP https://ift.tt/0bBRawH
from www.espn.com - TOP https://ift.tt/0bBRawH
Langganan:
Postingan (Atom)