Tech News

Today's Highlights

In case you missed it

Instructor creator Jason Liu asks developers why they still rely on ChatGPT over OpenAI Codex

+2 more

Top Tech Stories

Posts:95|Clusters:22|Next Crawl:--:--|

It’s a new day at Digg. Fresh stories are clustering now. 0% complete.

Recent Stars

Shopify/string_view Provides zero-copy read-only StringView into frozen Ruby strings via C extension for O(1) slicing without byte copies.
5h
#220
wedow/ticket Manages tickets as git-tracked markdown files with YAML frontmatter, supporting dependency graphs, priorities, and CLI queries in a single bash script.
5h839
#1015
alexei-led/ccgram Bridges Telegram to tmux/herdr windows to monitor and control Claude, Codex, Gemini CLI agents via keystrokes and output forwarding.
8h190
#1772
jstxn/agentdir Records coding-agent sessions in a local Maildir-style store for inspection, replay, search, and evidence auditing.
9h7
#220
tmux-plugins/tmux-resurrect Saves and restores tmux sessions, windows, panes, layouts and programs across restarts.
10h12.9k
#278

Github Stars

(7 days)

deepseek-ai/DeepSpec Provides full-stack code for training/evaluating speculative decoding draft models with data prep and eval scripts.
6d2.5k
#484

Privacy Policy|Terms of Service|© 2026 Digg Inc.

Privacy Policy|Terms of Service|© 2026 Digg Inc.

Yesterday's Top Stories•Jul 3, 2026

Posts: 3,995|Clusters: 860|Stories frozen at midnight PT

1

Instructor creator Jason Liu asks developers why they still rely on ChatGPT over OpenAI Codex

Engineer Andrew Mayne highlighted ATLAS as his primary reasoning tool

10h|8899152175.6k

2

T3 Stack creator Theo Browne asks how capable AI must get before developers stop reviewing its code

The prompt sparked debate over current trust in AI coding assistants.

5h|92.9k198356.1k

3

Trump administration adviser says he had to brief cabinet officials on dataset contamination and Roko's Basilisk

Other academics recalled explaining rationalist concepts like the Waluigi Effect to military officials

12h|1280316779.6k

4

Mistral AI releases Leanstral 1.5, a 119B MoE model for Lean 4 formal proof engineering

The open model achieved 100% on the miniF2F benchmark.

12h|1375126167.4k

5

Industry speculation points to an upcoming Thursday launch for OpenAI's rumored GPT-5.6 Sol model

The launch would precede Google DeepMind's rumored Gemini 3.5 Pro.

20h|63k207221.3k

6

The open-source pxpipe tool cuts Fable 5 token usage and costs by 60% by rendering text into images

The method exploits pricing differences between vision and text inputs

7h|6963717.9k

7

Outgoing White House adviser Sriram Krishnan says Trump administration won't establish centralized FDA-style AI regulator

Krishnan's duties included briefing officials on DeepSeek and explaining Roko's Basilisk.

12h|92123230.5k

8

Hamel Husain argues the need for forward-deployed AI engineers clashes with claims that AI replaces entire teams

Gary Marcus suggests this deployment gap disproves the existence of true AGI

12h|947013766.9k

9

Peter Thiel warns Vatican-backed global AI regulation efforts could undermine the US in its technology race with China

Thiel also predicted a democratic-socialist takeover of the Democratic Party.

22h|51k22492.1k

10

AI Security Institute says agent benchmarks must report performance curves over compute rather than static scores

A 10x increase in agent time horizon requires far more tokens.

23h|121382916.5k

11

Researchers distill Claude Fable 5 reasoning traces into Qwen3-4B, claiming zero output entropy and perfect self-consistency

The student model reportedly bypassed traditional teacher-performance limits.

6h|430516168.4k

12

Boom Supersonic founder Blake Scholl teases milestone post amid supportive reactions from tech investors

Jason Calacanis and Alexis Ohanian posted public congratulations.

12h|53.7k7649.2k

13

Chamath Palihapitiya says Tesla's reported $200 weekly AI spending cap suggests higher employee expenditures are likely wasteful

Commentators believe the limit sets a new industry budget benchmark.

21h|29.6k1.1k1.4M

14

Django co-creator Simon Willison routes agent tasks to cheaper subagents by letting Fable decide when to delegate

The autonomous routing experiment resulted in significant token savings.

11h|33.1k2.2k228.7k

15

X product head Nikita Bier shares an unsolicited DM pitching Elon Musk on a looksmaxing livestream

CEO Sam D'Amico offered to host the stream if Musk declines

5h|214.1k664978.2k

16

Policy scholar Dean W. Ball argues 2023 tech industry forecasts underestimated the steep trajectory of frontier AI capabilities

Critics prematurely predicted model capabilities would peak by 2030.

18h|458410858.6k

17

Meta CEO Mark Zuckerberg expresses disappointment in the industry's agentic AI progress, sparking debate among practitioners

Meta is reportedly preparing upgraded agentic coding capabilities for Meta AI

1d|348592204k

18

Mercor AI's Fable 5 model suffers a 10% performance regression on the APEX-SWE benchmark

The July update fell to 54.8% from June's 65.5%.

11h|45044874.7k

19

Signull labs creator limits replies to curb AI bots, prompting satirical mock error response from Josh Elman

The creator restricted replies to followed accounts to block automated spam

9h|38753057.1k

20

Andrew Gao, Cognition Labs AI builder, argues non-coders rarely hit Claude rate limits due to lower token consumption

Stability AI founder Emad Mostaque countered that philosophy and economics tasks regularly trigger his Claude limits

9h|4981016.4k

21

OpenAI's Gabriel Petersson demonstrates how attaching any random image dramatically boosts LinkedIn post reach

Bojan Tunguz plans to test the engagement hack next

19h|51.3k18196.3k

22

Mitchell Hashimoto, Ghostty creator, says frontier AI models have learned to accurately replicate his personal commit message style

The models absorbed two decades of his public open-source contributions

8h|22.6k22794.6k

23

Andy Jones questions whether AI safety reports are a consistent norm, citing declining page counts across recent models

He notes Gemini 3.5 has published no safety documentation.

12h|51481415.3k

24

OpenAI's roon argues automated post-training benchmarks are key to unlocking personalized AI models

GLM 5.2 topped the PostTrainBench at a fraction of competitors' costs

8h|284726566.1k

25

Researchers introduce COMPASS, a pan-cancer AI foundation model trained on 10,000 tumor samples to predict immunotherapy response

The model analyzes bulk tumor transcriptomes across 33 cancer types.

21h|340824478.1k

26

Software engineer Dan Luu says AI coding agents commit errors that would get human developers fired

He notes the industry paradox of using more agents to fix errors.

9h|433133.3k

27

Meta's upcoming Watermelon frontier model matches GPT-5.5 on internal benchmarks

The Muse variant consumes ten times more compute than Avocado.

1d|3782114124.6k

28

Judea Pearl argues AI labs make a mistake hiring philosophers, drawing pushback over computational philosophy contributions

Atoosa Kasirzadeh and Seth Lazar cited computational philosophy labs.

1d|41034816.3k

29

Prime Intellect research lead Will Brown demonstrates Fable generating a complete single-page HTML artifact in one attempt

Research engineer kache reacted to the rapid generation with laughter.

5h|37252254.4k

30

Books3 creator Shawn Presser receives $4,000 in cryptocurrency routed from a community meme coin

The financial support arrived while Presser faced housing insecurity.

23h|5213822k