Tech News

Today's Highlights

#1 last 24hrs

Google DeepMind releases DiffusionGemma, a 26-billion parameter MoE model for parallel, block-wise text generation

EL#686|@OMARSAR0

AC#324|@ANDREW_N_CARR

SD#97|@SEDIELEM

AK#65|@_AKHALIQ

SP#48|@SUNDARPICHAI

MOST NEGATIVE

Anthropic's undisclosed safeguards silently degrade Claude Fable 5 performance on AI self-improvement and frontier LLM research

CP#163|@CHAMATH

NL#134|@NATOLAMBERT

C🤗#58|@CLEMENTDELANGUE

#1 LIKED

9.5k

Péter Szilágyi, former Geth lead developer, warns Anthropic's Fable model blocking biology and cryptography foreshadows corporate gatekeeping

TR#1010|@TALIARINGER

VW#585|@VINCENTWEISSER

DF#388|@DANIELLEFONG

#1 VIEWED

OpenAI partners with Lionel Messi to launch "MessiMode" ChatGPT prompt for dyeing hair national flag colors

LA#844|@SCALING01

LK#487|@LUKASZKAISER

Top Stories

Posts:—|Clusters:—|Next Crawl:—

10

Anthropic's undisclosed safeguards silently degrade Claude Fable 5 performance on AI self-improvement and frontier LLM researchThe system uses steering vectors instead of explicit refusals.

11h|470.5k3.8k744

MOST NEGATIVE319

CE#1915|@INTUITMACHINE

TE#1489|@TENOBRUS

XL#1401|@XLR8HARDER

JS#1312|@JSUAREZ

CB#1269|@CODE_STAR

RS#1195|@ZIV_RAVID

AM#1047|@ANDREWDAI

SH#1026|@HAMANDCHEESE

TR#1010|@TALIARINGER

EL#896|@ELIEBAKOUCH

RP#847|@ROHANPAUL_AI

LA#844|@SCALING01

IR#814|@IRINARISH

GO#794|@GERGELYOROSZ

SH#708|@SHYAMALANADKAT

ZN#707|@ZACHARYNADO

EL#686|@OMARSAR0

DS#658|@DANSHIPPER

RT#596|@ROSSTAYLOR90

🇺🇦D#586|@DBAHDANAU

VW#585|@VINCENTWEISSER

EC#572|@ETHANCABALLERO

HP#571|@HYHIEU226

ML#528|@MISHALASKIN

AK#478|@BLACKHC

D🎇#457|@DAVIDAD

SS#444|@STEVESI

IO#443|@IANOSBAND

DF#388|@DANIELLEFONG

RA#386|@AGARWL_

SA#314|@PRFSANJEEVARORA

B(#269|@BEFFJEZOS

DR#223|@DELIPRAO

CP#163|@CHAMATH

NL#134|@NATOLAMBERT

C🤗#58|@CLEMENTDELANGUE

20

Google DeepMind releases DiffusionGemma, a 26-billion parameter MoE model for parallel, block-wise text generationIt runs four times faster on consumer GPUs

2h|480.6k8.1k#1BOOKMARKED2.2k#1COMMENTS683

NR#1858|@NATANIELRUIZG

KK#1592|@KASTNERKYLE

FB#1374|@XEOPHON

FO#1359|@FOFRAI

PS#1158|@_PHILSCHMID

ME#1056|@MERVENOYANN

PV#1054|@PETARV_93

CH#955|@KIMMONISMUS

RP#847|@ROHANPAUL_AI

BO#787|@BODONOGHUE85

DH#730|@DANIELHANCHEN

OS#699|@OSANSEVIERO

EL#686|@OMARSAR0

AC#324|@ANDREW_N_CARR

SD#97|@SEDIELEM

AK#65|@_AKHALIQ

SP#48|@SUNDARPICHAI

30

Biotechnology researchers claim Anthropic's safety filters render Claude unusable for potentially life-saving scientific researchThe Fable model is reportedly blocked from life sciences.

10h|253.6k2.7k412

CE#1915|@INTUITMACHINE

BR#1523|@BINDUREDDY

XL#1401|@XLR8HARDER

SD#1356|@INDUCTIONHEADS

TR#1010|@TALIARINGER

TA#738|@VICTORTAELIN

ZN#707|@ZACHARYNADO

EL#686|@OMARSAR0

AK#478|@BLACKHC

JC#401|@JULIEN_C

FF#396|@FRANCOISFLEURET

DF#388|@DANIELLEFONG

RS#305|@SCOBLEIZER

B(#269|@BEFFJEZOS

RS#70|@RICHARDSOCHER

C🤗#58|@CLEMENTDELANGUE

40

Socket finds malware targeting MCP developers bypasses AI scanners by triggering biosafety refusals with embedded nuclear textCompromised packages deploy JavaScript stealers using native extensions.

5h|534k6.3k2k

GL#1655|@GLEECH

ZE#1135|@ZEPHYR_Z9

TR#1010|@TALIARINGER

SK#621|@SEBKRIER

SS#444|@STEVESI

54

Geoffrey Irving, UK AI Security Institute chief scientist, launches Sequent Research to focus on superintelligence alignmentThe nonprofit plans to build automated alignment research tools

2h|55.2k864237

NI#1888|@NATHANPMYOUNG

LI#1847|@LIVGORTON

GL#1655|@GLEECH

TK#1611|@TOMEKKORBAK

JM#1111|@JESSEMHAN

GI#431|@GEOFFREYIRVING

DK#356|@DAVIDSKRUEGER

EP#290|@ETHANJPEREZ

61

Sam Altman says OpenAI may delay its IPO if recursive self-improvement occurs within six monthsOpenAI is also preparing a GPT-5.5 upgrade codenamed 5.6

2h|153.8k982199

SH#1026|@HAMANDCHEESE

CH#955|@KIMMONISMUS

SP#850|@STEPH_PALAZZOLO

LA#844|@SCALING01

ZN#707|@ZACHARYNADO

JL#629|@JESSICALESSIN

DP#437|@DANPRIMACK

AC#436|@ANDREWCURRAN_

70

River AI, led by @ibab, launches to build a fully user-owned personal AI stack to rival corporate modelsThe launch provides no technical details or release dates.

1h|270.2k811298

“P#1191|@PAULARAMBLES

EL#686|@OMARSAR0

HP#571|@HYHIEU226

FI#463|@FINBARRTIMBERS

RS#305|@SCOBLEIZER

GY#132|@THEGREGYANG

EM#88|@EMOSTAQUE

82

Péter Szilágyi, former Geth lead developer, warns Anthropic's Fable model blocking biology and cryptography foreshadows corporate gatekeepingTech founders and academics widely endorsed his concerns.

7h|422.4k#1LIKED9.5k897

TR#1010|@TALIARINGER

VW#585|@VINCENTWEISSER

DF#388|@DANIELLEFONG

91

Andreas Kirsch disputes claims that Anthropic is sabotaging its models, arguing performance changes are likely standard tuning adjustmentsExtropic's Guillaume Verdon originally amplified the sabotage claims.

10h|88.3k1k197

NI#1888|@NATHANPMYOUNG

VW#585|@VINCENTWEISSER

AK#478|@BLACKHC

DF#388|@DANIELLEFONG

B(#269|@BEFFJEZOS

BT#120|@BENTHOMPSON

103

OpenAI partners with Lionel Messi to launch "MessiMode" ChatGPT prompt for dyeing hair national flag colorsUsers upload selfies and apply a custom image-editing prompt

2h|#1VIEWS750.4k7k801

LA#844|@SCALING01

LK#487|@LUKASZKAISER

117

The Pragmatic Engineer's Gergely Orosz warns Anthropic's Fable retains prompts for 30 days and can unilaterally downgrade modelsUsers cannot opt out of the 30-day storage policy.

10h|299.6k5.6k436

CE#1915|@INTUITMACHINE

GO#794|@GERGELYOROSZ

124

Erik Brynjolfsson's Stanford lab launches AI Economic Indicators to track macroeconomic impacts of AIThree dashboards track real-world adoption and deployment patterns.

3h|58.1k504245

CW#1604|@CALEBWATNEY

RI#733|@RISHIBOMMASANI

EB#333|@ERIKBRYN

EM#117|@EMOLLICK

136

CoreAutoAI co-founder Rohan Anil shows targeted tuning lets Meta's vanilla Distributed Shampoo achieve competitive NanoGPT speed-run performanceThe benchmark compared performance against an obsolete Muon baseline.

9h|61.9k523161

BI#1872|@BILALTWOVEC

YJ#1212|@YOUJIACHENG

EL#896|@ELIEBAKOUCH

ZN#707|@ZACHARYNADO

KJ#468|@KELLERJORDAN0

RA#123|@_AROHAN_

141

Former xAI co-founder Igor Babuschkin launches River AI to build user-owned personal AI systemsThe startup plans to begin shipping products soon.

1h|64k594143

EL#896|@ELIEBAKOUCH

BA#693|@BAYESLORD

RS#305|@SCOBLEIZER

AG#284|@ALEXGRAVELEY

153

Sriram Krishnan argues open scientific research faces a collision course with rapid AI singularity transitionsYann LeCun noted the factions have clashed for years.

8h|55.2k51418

CF#1944|@CFGEEK

EL#686|@OMARSAR0

VW#585|@VINCENTWEISSER

ML#528|@MISHALASKIN

GT#44|@GARRYTAN

165

Anthropic's Fable 5 model migrates a 50-million-line Ruby codebase in one day on Cognition's FrontierCode benchmarkExperts debate whether its performance gains over GPT-5.5 are incremental.

10h|67.6k82576

FF#396|@FRANCOISFLEURET

T(#346|@TEORTAXESTEX

AG#284|@ALEXGRAVELEY

KA#229|@YACINEMTB

MR#222|@MAITHRA_RAGHU

175

Nathan Lambert, AI2 post-training lead, says Anthropic abruptly cut off his access to its advanced models without warningOther technical operators reported similar sudden Claude access cuts

4h|65.8k1.8k108

DP#476|@DIMITRISPAPAIL

FI#463|@FINBARRTIMBERS

NL#134|@NATOLAMBERT

185

Dylan Patel says Anthropic's API token share fell to 73% as Mythos 5 refusals drove SemiAnalysis testers to OpenAI's CodexThe shift highlights low stickiness in AI infrastructure markets.

2h|80.9k951131

GO#794|@GERGELYOROSZ

AM#759|@ANDREWMAYNE

DF#388|@DANIELLEFONG

DP#83|@DYLAN522P

1913

Anthropic launches Claude Fable 5, routing flagged sensitive queries to an Opus 4.8 fallback modelThe model is optimized for software engineering and autonomous tasks.

8h|14.6k11813

RP#847|@ROHANPAUL_AI

LA#844|@SCALING01

EL#686|@OMARSAR0

DF#388|@DANIELLEFONG

209

Developer ChrissGPT used Anthropic's Claude Fable to generate an 8,000-line Gen-1 Pokémon game in a single promptThe generation required one hour of model reasoning.

9h|74.5k30598

TR#1010|@TALIARINGER

CH#955|@KIMMONISMUS

RP#847|@ROHANPAUL_AI

DE#158|@DEEDYDAS

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